Review Reports - Design and Testing of a Vision-Based, Electrically Actuated, Row-Guided Inter-Row Cultivator

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

I. General Comments on the Manuscript
This manuscript describes the development and testing of an intelligent system for inter-row cultivation of maize, which uses a vision system (YOLOv8s) and an electrically actuated lateral displacement mechanism. The proposed system employs plant recognition and a row-matching algorithm to calculate the lateral offset in real time and, in combination with delay compensation, controls a servo-electric cylinder for precise positioning of the cultivator.
The topic is current and fully relevant to the field of precision agriculture and automated inter-row cultivation. It falls entirely within the scope of Agronomy and is of interest to a wide range of specialists working in agricultural technology and sustainable management aimed at reducing the use of chemicals through mechanical weeding. The manuscript is suitable for publication after revisions focused mainly on clarifying the experimental conditions, correcting figures and formulas, and expanding the discussion on the system’s limitations, as outlined below.
II. Specific Comments
To improve the manuscript, the following remarks and recommendations are addressed to the authors, organized by section:
1. Introduction – The introduction is logically structured and provides a comprehensive review of the relevant literature. The authors adequately describe the limitations of existing approaches, but a more clearly formulated research gap is needed to specify what exactly remains unresolved in the literature.
1.1. I recommend adding one or two sentences that clearly summarize what is missing in the current methods for vision-based control of inter-row cultivators (e.g., delay compensation, limited lateral correction, or quantitative assessment of root damage risk).
1.2. Additionally, the authors are encouraged to include one or two sentences that more explicitly distinguish the advantages of electric actuation (used here) compared to hydraulic systems, which are often employed in larger machines. Why was an electric cylinder chosen — for its accuracy, response speed, or environmental benefits?
2. Materials and Methods
This section is overall very strong, well-structured, and demonstrates a thorough engineering approach to addressing critical challenges in precision agriculture—particularly regarding delay compensation and quantitative risk assessment. However, several important clarifications are missing and should be added or revised to make the section complete and fully understandable to readers:
2.1. In Section 2.2.1, it is mentioned that a YOLOv8s-based model was used for segmentation and that images were collected and annotated. The authors are encouraged to add a short paragraph or a table describing the dataset, including the capture conditions and which specific classes were validated (e.g., maize only, or also other plants/weeds) for the training of YOLOv8s-Seg.
2.2. The methodology for experimentally calibrating the time delay—from image processing on the Jetson Orin to command transmission to the C37 controller—should be described in detail.
2.3. Please confirm whether the diameters of the root zones (10, 15, 20 cm) were empirically measured in the field or adopted from literature sources.
2.4. I address the following essential recommendation regarding the notation of symbols (d, D, g) in formulas (1), (2), and (3): In mechanical engineering, the symbols D and d are traditionally reserved for diameters. Their use for other variables (as in formula (1), where D and d represent distances) may lead to confusion when interpreting the equations. Alternative symbols are recommended.
This also applies to the symbol g, which is conventionally used to denote gravitational acceleration (≈9.81 m/s²) and should not be used for other quantities.
2.5. Please revise the presentation of all variables below formulas (1), (2), and (3) by adding standard measurement units in parentheses after each symbol (e.g., m, s, m/s; V is the speed of the implement (m/s)). This is a standard requirement in engineering publications and is essential for improving clarity and reproducibility of your methods.
2.6. The authors should check and correct the figure numbering in the manuscript. There is a duplication in the citation of ―Figure 5‖ (lines 247 and 291). Each figure must have a unique, sequential number corresponding to its first appearance in the text.
2.7. For greater visual clarity, it is recommended that figures be cleaned of embedded text and labels. Instead, components should be marked with numbered positions (1, 2, 3, …), and the full description of these components should be moved to the figure caption/legend below each image.
2.8. In the field tests, essential information is missing, such as: number of replications, duration of each pass, average operating speed, weather conditions, soil parameters, and how root damage was assessed (manual scoring? high-resolution inspection?). The authors are encouraged to provide this information.
2.9. The authors should consider adding a new subsection (e.g., 2.4 Statistical Analysis), clearly stating the software used to calculate the statistical metrics (MAE, RMSE, SD, E95) and whether a significance level was applied in the analysis of results.
3. Results and Discussion
The Results and Discussion sections are clearly structured, well linked to the methodology, and provide a convincing quantitative and qualitative interpretation of the results obtained from the laboratory and field experiments. The following recommendations are addressed to the authors:
3.1. The current title of the section is ―3. Results.‖ It is recommended to change it to ―3. Results and Discussion‖ to better reflect the integrated nature of the section, where comparisons with previous studies are made and conclusions are drawn.
3.2. Revise the Discussion subsection by shortening and refocusing it. Remove the repetition of numerical values from Table 2 and concentrate instead on the analysis, interpretation, and comparison of your scientific findings with previous research, emphasizing the agronomic relevance of the obtained results.
3.3. It is recommended to include a statistical significance analysis (e.g., ANOVA or t-test) to compare the mean root-damage risk values across different growth stages. Presenting the differences (0.12%, 1.46%, 9.61%) as absolute facts without statistical validation reduces the scientific weight of the conclusions.
4. Conclusion
The Conclusion section fulfills its main purpose — to summarize the results and confirm the hypothesis that electric actuation can provide effective and safe inter-row cultivation. The following recommendations are addressed to the authors:
4.1. Add one sentence that clearly summarizes the main contribution of the proposed system compared with previous developments.
4.2. Include a brief paragraph outlining future research directions, such as testing at higher operating speeds, optimizing the control algorithm, and extending the system’s applicability to other crops.
5. References
The references used are relevant and well selected. Most of the cited publications are from the last five to seven years, which is entirely appropriate for the field of precision agriculture, vision systems, and intelligent control of agricultural machinery.

Author Response

Design and Testing of a Vision-Based, Electrically Actuated Row-Guided Inter-Row Cultivator

Dear Reviewer,

On behalf of all the authors, we sincerely appreciate your valuable comments on the manuscript. Your comments not only provided constructive suggestions for improving the quality of the manuscript but also led us to consider our approaches and the design of the system in detail. These comments will also promote our future research.

Best regards,

Haonan Yang, Liyan Wu*, Changyuan Zhai*

Comment # 1.1: I recommend adding one or two sentences that clearly summarize what is missing in the current methods for vision-based control of inter-row cultivators (e.g., delay compensation, limited lateral correction, or quantitative assessment of root damage risk).

Response to Comment 1.1:: Thank you for this helpful suggestion. We agree that the limitations of current vision-based inter-row cultivators should be summarized more explicitly. To address this, we have revised the end of the Introduction and inserted the following sentences, which now clearly state what is still missing in existing methods

Lines:108-115

Most existing vision-based inter-row guidance systems do not explicitly model or compensate the perception–control–actuation delay along the forward direction; their lateral correction stroke and dynamic bandwidth are seldom evaluated over the full operating range, and quantitative, spatially explicit assessment of root-damage risk along the crop row is almost never reported.As a result, even when centimetre-level tracking errors are achieved, it remains unclear how much of the root zone is actually endangered at different growth stages and how early inter-row cultivation should be scheduled to keep the risk within an acceptable range.

Comment # 1.2: Additionally, the authors are encouraged to include one or two sentences that more explicitly distinguish the advantages of electric actuation (used here) compared to hydraulic systems, which are often employed in larger machines. Why was an electric cylinder chosen — for its accuracy, response speed, or environmental benefits?

Response to Comment 1.2: Thank you for raising this point. We agree that the advantages of electric actuation over hydraulic systems should be stated more explicitly. In the revised Introduction, we now explain why an electric cylinder was selected for the proposed inter-row cultivator. Specifically, we added the following sentences in the paragraph that introduces the electrically actuated row-guided cultivator and control framework:

Lines:118-124

In the proposed implement, a servo-electric cylinder provides lateral shifting of the cultivator relative to the crop rows. Compared with hydraulic lateral-shift systems widely used on larger inter-row cultivators, this electric actuator offers finer position resolution and repeatability, a more compact drive with no risk of oil leakage, and straightforward integration with electronic controllers and delay-compensation algorithms, making it particularly suitable for achieving centimetre-level lateral tracking on a low-speed, medium-width maize cultivator.

Comment # 2.1: In Section 2.2.1, it is mentioned that a YOLOv8s-based model was used for segmentation and that images were collected and annotated. The authors are encouraged to add a short paragraph or a table describing the dataset, including the capture conditions and which specific classes were validated (e.g., maize only, or also other plants/weeds) for the training of YOLOv8s-Seg.

Response to Comment 2.1: Thank you for this helpful suggestion. We agree that the dataset used for training and validating the YOLOv8s-Seg model should be described more clearly. In the revised Section 2.2.1, we have added a detailed description of the image acquisition conditions, the dataset composition, and the annotation procedure, as well as the performance of the trained model.

Lines:196-203

The maize seedling images used in the experiments were acquired at the National Precision Agriculture Research and Demonstration Base in Xiaotangshan, Changping District, Beijing, from 1 July to 20 July 2025, between 09:00 and 16:00. The images were captured using a Logitech C930c camera at a resolution of 1920 × 1080. To ensure background complexity and dataset diversity, images were collected under various weather conditions, illumination levels, and growth stages, yielding a total of 1,987 images. Maize seedlings in the dataset were annotated as target objects using the LabelImg software, and the images were divided into training, validation, and test sets in a ratio of 8:1:1.

Lines:205-208

The model trained in this study is a single-class detector targeting maize seedlings. After 100 training epochs, it achieved a precision (P), recall (R), and mean average precision (mAP) of 96.7%, 95.8%, and 96.2%, respectively. Crop-row detection was performed using a segmentation model based on YOLOv8s.

Comment # 2.2:The methodology for experimentally calibrating the time delay—from image processing on the Jetson Orin to command transmission to the C37 controller—should be described in detail.

Response to Comment 2.2: Thank you for this helpful comment. We agree that the procedure for determining the overall delay between image processing on the Jetson Orin NX and command transmission to the C37 controller should be described more clearly. In the revised Section 2.2, we have added a concise description of an end-to-end timing test of the communication chain to explain how this perception–communication delay was experimentally calibrated.

Lines:223-234

To characterize the overall delay between the completion of image processing on the Jetson Orin NX and the transmission of the command to the C37 controller, we performed an end-to-end timing test of the communication chain. In this test, the vision program recorded a timestamp immediately after finishing YOLOv8s-Seg inference and computing the lateral offset on the Orin NX. The communication thread then packaged the offset into a fixed 16-byte row-guidance frame and sent it via the USB–CAN interface. A second timestamp was recorded when this frame was actually placed on the CAN bus and accepted by the interface. This procedure was repeated over multiple control cycles under typical operating conditions, and the average difference between the two timestamps was taken as the overall perception–communication delay from the Orin NX to the C37. This experimentally calibrated delay value was then used as the perception–communication component of the total system delay in the delay-compensation model.

Comment # 2.3: Please confirm whether the diameters of the root zones (10, 15, 20 cm) were empirically measured in the field or adopted from literature sources.

Response to Comment 2.3: Thank you for this important question. In this study, the root-zone planar diameters were adopted from published measurements of the lateral spread of maize roots under comparable cultivation conditions. In the revised Section 2.3, we now explicitly state that the root-zone diameters at each growth stage are taken from the results reported in [29,30].

Lines:372-375

The root-system plan-view diameters at each growth stage were selected based on measurements of the lateral spread of maize roots under comparable cultivation conditions reported in [29,30].

Comment # 2.4: I address the following essential recommendation regarding the notation of symbols (d, D, g) in formulas (1), (2), and (3): In mechanical engineering, the symbols D and d are traditionally reserved for diameters. Their use for other variables (as in formula (1), where D and d represent distances) may lead to confusion when interpreting the equations. Alternative symbols are recommended.

Response to Comment 2.4: Thank you for this essential recommendation regarding the notation of the symbols in formulas (1)–(3). We agree that using D and d for distances and g for a non-gravitational quantity may cause confusion for readers with a mechanical engineering background. In the revised manuscript, we have therefore replaced D and d in formula (1) with and , which now denote the delay-compensation distance and the forward distance from the camera field-of-view centre to the blade, respectively. In formulas (2) and (3), the symbol g has been replaced by , which denotes the travel distance per pulse. All corresponding variable descriptions below the formulas and all occurrences in the text have been updated accordingly to ensure consistent and unambiguous notation throughout the manuscript.

Lines:271-273

(1)

Where is the delay compensation distance in meters, is the distance from the field-of-view center to the cultivator blade in meters, is the real-time constant speed of the machine in meters per second, and is the total system delay time in seconds.

Comment # 2.5: Please revise the presentation of all variables below formulas (1), (2), and (3) by adding standard measurement units in parentheses after each symbol (e.g., m, s, m/s; V is the speed of the implement (m/s)). This is a standard requirement in engineering publications and is essential for improving clarity and reproducibility of your methods.

Response to Comment 2.5: Thank you for this practical suggestion. In the revised manuscript, we have updated the presentation of all variables below formulas (1), (2), and (3) by adding the corresponding standard measurement units in parentheses after each symbol. For example, in formula (1) we now specify as the delay-compensation distance (m), as the forward distance from the camera field-of-view centre to the blade (m), as the implement speed (m/s), and as the total system delay (s). In formulas (2) and (3), we explicitly define as the travel speed of the implement (m/s), as the pulse frequency (Hz), as the travel distance per pulse of the implement (m/pulse), as the travel distance of the implement (m), and and as the counter value and pulse count, respectively. We have checked that all variable descriptions in this part now include units in a consistent and unambiguous way.

Lines:271-273

(1)

Lines:288-290

	(2)
	(3)

is the travel speed of the implement (m/s); is the pulse frequency (Hz); is the travel distance per pulse of the implement (m/pulse); is the travel distance of the implement (m); is the counter value; and is the pulse count.

Comment # 2.6: The authors should check and correct the figure numbering in the manuscript. There is a duplication in the citation of ―Figure 5‖ (lines 247 and 291). Each figure must have a unique, sequential number corresponding to its first appearance in the text

Response to Comment 2.6: Thank you for pointing out the problem with the figure numbering. In the revised manuscript, we have carefully checked all figures and their in-text citations and corrected the duplicated reference to “Figure 5” (previously cited twice around lines 247 and 291). Each figure now has a unique, sequential number that corresponds to its first appearance in the text, and all subsequent citations have been updated accordingly to ensure consistency.

Lines:357

Figure 6. (a) Schematic of the controlled-site experiment；(b) Sampling layout schematic.

Comment # 2.7: For greater visual clarity, it is recommended that figures be cleaned of embedded text and labels. Instead, components should be marked with numbered positions (1, 2, 3, …), and the full description of these components should be moved to the figure caption/legend below each image.

Response to Comment 2.7: Thank you for this valuable suggestion. To improve the visual clarity and readability of the figures, we have revised the schematic figures that previously contained embedded text and labels. In the revised manuscript, the main components in these figures are now indicated by numbered callouts (1, 2, 3, …) within the figure panels, and the full names and descriptions of these components have been moved to the corresponding figure captions/legends below each image. We have also checked that any remaining text inside the figures is kept to a minimum and limited to essential symbols or axes labels, in accordance with the journal’s style recommendations.

Lines:274-278

Figure 4. Schematic Diagram of the Control System Delay. (a) Network camera; (b) NVIDIA Jetson Orin NX; (c) C37 Controller; (d) Servo drive; (e) Servo electric cylinder; 1. Cabbage identification and positioning time; 2. Communication time; 3. Servo driver response time; 4. Opening and closing time.

Comment # 2.8: In the field tests, essential information is missing, such as: number of replications, duration of each pass, average operating speed, weather conditions, soil parameters, and how root damage was assessed (manual scoring? high-resolution inspection?). The authors are encouraged to provide this information.

Response to Comment 2.8: Thank you for pointing out that the description of the field tests lacked important information on the experimental conditions and evaluation procedure. In the revised Materials and Methods, Section 2.3.2 (Field Experiments) has been expanded. We now state that, for each growth stage, three replicate row-guided cultivation passes were carried out on a 30 m effective row segment, that the duration of a single pass was calculated from the RTK trajectory timestamps (average 61 s), and that the actual operating speed was a constant 0.51 m/s. We have also clarified that meteorological data during the three cultivation stages were obtained from the automatic weather station at the experimental base; for each cultivation date, daily mean air temperature, mean relative humidity, 10-min mean wind speed, and daily global solar radiation were extracted, and the values are summarized in Table 1. In addition, we have added a detailed description of the root-injury assessment method: the sowing and cultivation RTK trajectories are projected into a unified coordinate system, a circular root-influence zone is defined for each plant based on the root-system planar diameters reported in [29,30], and root damage is quantified as the proportion of the cultivation trajectory whose discrete RTK points fall within this zone. These additions make the field-test description more complete and improve the reproducibility and interpretability of the root-damage risk analysis.

Lines:381-408

To verify the root-injury risk of the row-guided inter-row cultivator under field conditions, a maize trial was conducted at the Xiaotangshan National Precision Agriculture Research Demonstration Base. During seeding, an RTK receiver was mounted at the end of the planter crossbeam and aligned longitudinally with the metering unit; the planting-row trajectories were continuously recorded in a unified projected coordinate system, and the ground-projected RTK data were used to define reference line a (Figure 7, left). During inter-row cultivation, an RTK receiver of the same model was mounted at the geometric midpoint of the cultivator crossbeam to obtain the implement-center trajectory. Based on the row spacing of 50 cm, this trajectory was shifted 25 cm to the left in the same coordinate system to form the target inter-row cultivation path, which was then fitted to the sowing trajectory so that cultivation was performed on the same crop row (Figure 7, right). The target operating speed was 0.5 m/s, and the actual operating speed, back-calculated from the RTK timestamps, was a constant 0.51 m/s. The plot parameters were: row spacing 50 cm, in-row plant spacing 25 cm, and an effective single-row length of 30 m; buffer sections of 5 m at both ends of each row were reserved for acceleration and steady running [28], and data from these sections were excluded. To examine the influence of growth stage on risk, three sets of parameters were defined: 2-3- leaf stage with tillage depth 5 cm and root-system plan-view diameter 10 cm; 4-5-leaf stage with tillage depth 15 cm and root-system plan-view diameter 15 cm; and 6-8-leaf stage with tillage depth 10 cm and root-system plan-view diameter 22 cm. The root-system plan-view diameters for each growth stage were selected according to measurements of the lateral spread of maize roots under comparable cultivation conditions reported in [29,30]. For each growth stage, field inter-row cultivation tests were repeated three times, each covering a 30 m effective working section, and the duration of a single pass was calculated from the RTK trajectory timestamps, giving an average of 61 s. By keeping the seeding and cultivation trajectories coincident and using the same reference line a, spatial consistency between the two stages was ensured, allowing the root-damage rate to be quantitatively related to geometric deviation within a unified coordinate frame.

Lines:412-439

Meteorological data during the experiments were obtained from the automatic weather station at the experimental base. For each inter-row cultivation operation at the 2–3-, 4–5- and 6–8-leaf stages, daily mean air temperature, mean relative humidity, 10-min mean wind speed and daily global solar radiation on the corresponding date were extracted, and the statistics are summarized in Table 1. The latitude–longitude data of the sowing and cultivation trajectories were fitted into geometric strips, and a strip representation consisting of the root-distribution zone, the cultivation zone and the root-injury zone was constructed; the intersections of these strips were used as the basis for judging root injury. According to the root-system plan-view diameters at the 2–3-, 4–5- and 6–8-leaf stages reported in [29,30], a circular root-influence zone was defined for each plant by extending half of the corresponding diameter from the sowing-row center line to either side, and this circle was regarded as the effective root-activity area of a single maize plant. After projecting the discrete RTK coordinates of the cultivation trajectory into the same coordinate system, the lateral position of each point was examined to determine whether it fell within the root-influence zone of the corresponding crop row; if it did, the cultivating blade at that location was considered to have entered the root zone and the point was counted as a root-injury point. Using the fixed spacing of the common coordinate axis together with the change in latitude between neighboring positions, the forward travel length associated with each location entering the root zone was calculated. The forward lengths of all points judged to have entered the root zone were then accumulated to obtain the total length inside the root zone. Using the same procedure over the entire operation yielded the total arc length of the cultivation trajectory. The root-damage rate was defined as the ratio of the total length inside the root zone to the total arc length of the cultivation trajectory, and the mean value of the three replicate operations was used to characterize the root-damage risk at each growth stage. It should be emphasized that only the overlapping segment between the cultivation band and the safety (root-distribution) zone contributes to the root-injury arc length; therefore, the root-damage rate actually reflects the proportion of this overlapping segment in the total cultivation trajectory length.

Comment # 2.9: The authors should consider adding a new subsection (e.g., 2.4 Statistical Analysis), clearly stating the software used to calculate the statistical metrics (MAE, RMSE, SD, E95) and whether a significance level was applied in the analysis of results.

Response to Comment 2.9: Thank you for this helpful suggestion. In the revised Materials and Methods, we have added a new subsection entitled “2.4 Statistical Analysis”. In this subsection, we specify that the statistical metrics (MAE, RMSE, SD, E95) were calculated in Microsoft Excel based on the original measurement data, and we describe how these indices were obtained from the error sequences. We also state that, when comparing the mean root-damage rates among different growth stages, one-way ANOVA was performed in Excel with a significance level of p < 0.05. These additions clarify the data-processing procedure and the statistical assumptions underlying the reported results.

Lines:441-459

2.4 Statistical Analysis

The statistical metrics (MAE, RMSE, SD and E95) were calculated from the original measurement data. For each cultivation pass, the lateral tracking error between the implement trajectory and the target row-guidance trajectory was obtained as an error sequence consisting of all discrete sampling points along the path. This error sequence was processed in Microsoft Excel. The mean absolute error (MAE) was computed as the arithmetic mean of the absolute errors, the root mean square error (RMSE) was obtained as the square root of the mean of the squared errors, the standard deviation (SD) was calculated from the error sequence, and the 95th percentile error (E95) was defined as the 95th percentile of the absolute error values. For each operating condition, the reported statistics are the averages over three replicate passes and are used to characterize the row-tracking performance under that condition.
In the root-damage risk analysis, the root-damage rate for a single pass was calculated as the ratio of the root-injury arc length (length of the trajectory segment overlapping the root-influence zone) to the total cultivation arc length, and the mean over three replicate passes was taken as the root-damage index for each growth stage. To assess whether the differences in mean root-damage rate among growth stages were statistically significant, one-way analysis of variance (ANOVA) was performed in Excel for the 2–3-, 4–5- and 6–8-leaf stages, with the significance level set at p < 0.05.

Comment # 3.1: The current title of the section is ―3. Results.‖ It is recommended to change it to ―3. Results and Discussion‖ to better reflect the integrated nature of the section, where comparisons with previous studies are made and conclusions are drawn

Response to Comment 3.1: Thank you for this helpful suggestion. We agree that the current section integrates both presentation of results and their interpretation. In the revised manuscript, we have changed the heading “3. Results” to “3. Results and Discussion” to better reflect the combined nature of this section and to align with standard scientific writing practice.

Lines:460

Results and Discussion

Comment # 3.2: Revise the Discussion subsection by shortening and refocusing it. Remove the repetition of numerical values from Table 2 and concentrate instead on the analysis, interpretation, and comparison of your scientific findings with previous research, emphasizing the agronomic relevance of the obtained results.

Response to Comment 3.2: Thank you for this helpful suggestion. We agree that the Discussion section should focus more on analysis and interpretation of the results rather than repeating the numerical values from the tables. Accordingly, in the revised manuscript we have shortened and reorganized Section 4 Discussion. The repeated description of the numerical results in Table 3 has been removed, and the revised paragraphs now place greater emphasis on: (1) explaining the mechanisms behind the stage-dependent patterns of root-damage risk; (2) comparing our findings with previous studies on camera-guided hoeing and near-crop in-row tool control; and (3) highlighting the agronomic implications for selecting cultivation timing and operating parameters. We believe these changes make the Discussion more concise and focused, and more consistent with standard scientific writing practice.

Lines:611-658

Within a unified coordinate framework, this study registers the row-guidance trajectories, root distribution, and cultivation outcomes. The results show that as the growth stage advances, the inter-row safety belt continuously shrinks and tends to close; even under centimeter-level row-guidance error, multiple local and continuous overlaps still occur in the middle and late stages, causing the injury-arc length and root-damage rate to increase in step with the lateral expansion of the root system. The field results can be summarized in three stages: at the 2–3-leaf stage, root damage remains at a very low level while weed control can be maintained; at the 4–5-leaf stage, a compromise is achieved between safety and coverage; at the 6–8-leaf stage, pronounced accumulation of root damage appears under the same tracking accuracy. This indicates that cultivation timing is the primary risk-control variable under given row spacing and operational accuracy, whereas geometric row-guidance error alone cannot fully represent operational safety. Repeated overlaps at headlands, outer rows, and segments with abrupt changes in row direction suggest that sowing-row skew and micro-topographic undulations amplify local risk, but the overall trend is still dominated by the temporal lateral expansion of the root zone. Taken together, the spatial-registration evidence indicates that early cultivation provides a larger safety margin; the 4–5-leaf stage is the recommended window that balances root safety and weed-control coverage; and if cultivation is required at the 6–8-leaf stage for management reasons, shallow tillage, reduced working width, and moderate speed reduction should be adopted to lower the probability of entering the root zone while retaining the necessary coverage.

Compared with existing camera-guided hoeing studies, this work introduces a quantitative evaluation of root damage under a similar level of row-guidance accuracy. Previous studies have achieved high inter-row operational efficiency and stable geometric accuracy at low to medium speeds, but system performance has mostly been evaluated by indicators such as row-center deviation, and comparable root-damage data are rarely reported [35]. Research in near-crop spaces has shown that, for crops such as tomato and lettuce, geometric detection and real-time control of in-row tools can realize deterministic trajectories within a narrow safety belt, whereas higher speeds or unstable implement attitude tend to amplify lateral deviation and increase the risk of crop interference [36]. In contrast, the present study uses electric lateral shifting and delay compensation to keep row-guidance error within the centimeter range and, at this accuracy level, achieves a combined performance of low root damage and relatively high coverage. Thus, the performance evaluation is extended from geometric accuracy to root-risk indicators, providing experimental evidence for determining maize cultivation windows under root-zone safety constraints.

From an engineering perspective, the system takes the sowing RTK trajectory as the primary reference, uses visual detection for auxiliary correction, and combines delay compensation with servo electric-cylinder control so that the row-position information acquired during sowing can be used directly for row guidance during cultivation. This maintains consistency between sowing and cultivation tracks in fields with terrain undulations and row-direction deviations. Field results show that row-guidance error increases with forward speed, while the effective width of the safety belt is correspondingly reduced and local risk increases [38]; related studies have likewise indicated that tight coupling among perception, positioning, and actuation is essential for maintaining trajectory consistency under continuous operation [39,40]. Therefore, combining low-delay control and trajectory-consistency design with parameter adaptation offers promising engineering potential for wider deployment of such systems.

Comment # 3.3:It is recommended to include a statistical significance analysis (e.g., ANOVA or t-test) to compare the mean root-damage risk values across different growth stages. Presenting the differences (0.12%, 1.46%, 9.61%) as absolute facts without statistical validation reduces the scientific weight of the conclusions.

Response to Comment 3.3: Thank you very much for this important comment. In our original data processing, the root-damage rates at the 2–3-, 4–5- and 6–8-leaf stages were already evaluated using one-way analysis of variance (ANOVA), but this procedure and its outcome were not explicitly described in the manuscript. In the revised version, we have clarified this in Section 2.4 “Statistical Analysis”, where we state that the differences in mean root-damage rate among growth stages were tested by one-way ANOVA in Excel, with the significance level set at p < 0.05. In addition, the field experiment results in the subsection around Figure 14 and Table 3 have been updated to explicitly report the ANOVA findings: based on the data from three replicate passes, one-way ANOVA showed that growth stage had a highly significant effect on root-damage rate (p < 0.001); the mean root-damage rate at the 6–8-leaf stage was significantly higher than at the 2–3- and 4–5-leaf stages, whereas the difference between the 2–3- and 4–5-leaf stages was not statistically significant. These revisions provide a formal statistical basis for the reported values (0.12%, 1.46%, 9.61%) and strengthen the scientific validity of the conclusions.

Lines:441-459

2.4 Statistical Analysis

Lines:

Based on the data from three replicate passes, the results of one-way analysis of variance (ANOVA) showed that growth stage had a highly significant effect on root-damage rate (p < 0.001); the mean root-damage rate at the 6–8-leaf stage was significantly higher than at the 2–3- and 4–5-leaf stages, whereas the difference between the 2–3- and 4–5-leaf stages was not statistically significant.

Comment # 4.1: Add one sentence that clearly summarizes the main contribution of the proposed system compared with previous developments.

Response to Comment 4.1: Thank you for this valuable suggestion. In the revised manuscript, we have added one sentence at the end of the first paragraph of the Conclusion to clearly summarize the main innovation and contribution of the proposed system compared with previous related studies and equipment. Specifically, the new sentence states that, relative to systems that mainly rely on hydraulic actuation or evaluate performance only by geometric deviation, our system integrates electric lateral shifting with delay compensation to maintain centimeter-level row-guidance accuracy while introducing injury-arc–based risk quantification and cultivation-window determination, thus providing an engineering-feasible route for safe mechanical inter-row cultivation in maize at the seedling stage.

Lines:672-678

Compared with existing row-guided cultivators that mainly rely on hydraulic actuation or evaluate performance only in terms of geometric deviation, the proposed system integrates electric lateral shifting with delay compensation to maintain centimeter-level row-guidance accuracy while introducing injury-arc–based risk quantification and cultivation-window determination, thereby providing an engineering-feasible route for safe mechanical inter-row cultivation of maize at the seedling stage.

Comment # 4.2: Include a brief paragraph outlining future research directions, such as testing at higher operating speeds, optimizing the control algorithm, and extending the system’s applicability to other crops.

Response to Comment 4.2: Thank you for this helpful suggestion. We agree that the Conclusion should briefly indicate the directions for future research. In the revised manuscript, we have added a concise closing sentence at the end of the Conclusion that outlines our planned work, including strengthening force/attitude feedback and injury-risk–based closed-loop control, and evaluating the adaptability of the proposed system to other row crops and cultivator configurations. The detailed changes are reflected in the revised Conclusion section.

Lines:680-683

Future work will focus on refining force and attitude feedback and injury-risk–based closed-loop control, and on validating the system’s adaptability to other row crops and cultivator configurations.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors * What is the main question addressed by the research? The main question addressed by the research: How to develop an intelligent vision-based row-guided inter-row cultivator with electric lateral shifting? * Do you consider the topic original or relevant to the field? Does it address a specific gap in the field? Please also explain why this is/ is not the case. I consider that this topic is original and relevant to the field of mechanical weeding processes. These processes prohibit the use of chemicals in cultivation, which means a decrease in chemical use in agriculture, less cumulative CO2 emissions from agriculture in total, and higher biodiversity (less chemicals - less damage to biodiversity) in agricultural fields. * What does it add to the subject area compared with other published material? It adds a complete method of weeder design, experiment planning, experiment execution, and results elaboration for this original machine. * What specific improvements should the authors consider regarding the methodology? The author presented a complete methodology for this research. They do not need to add something new. * Are the conclusions consistent with the evidence and arguments presented, and do they address the main question posed? Please also explain why this is/is not the case. The conclusions are consistent with the evidence and arguments presented address the main question posed. There is described a vision-based row-guidance inter-row cultivator for maize is described. It integrates a deep-learning row-line detection module with a lateral-shift servo-servo-electric-cylinder actuator, together with a root-injury quantification method based on a unified coordinate frame and arc-length integration. There are also experimental results presented for controlled-site offset with the Mean Absolute Error (MAE), the root-injury rate with the coverage at the 4-5 leaf stage, and at the 6-8 leaf stage. All data are discussed further. They point out that: - The root damage risk grows together with the plant's growth. - The bigger the tool shift effects, the bigger the loads of machinery. - The actuators' delay compensation is needed for accurate machinery steering. - The optimal weeding of the maize crop is when the maize has 4-5 leaves. - When maize has more than 5 leaves, more elastic and shallow tools are needed.

Author Response

Design and Testing of a Vision-Based, Electrically Actuated Row-Guided Inter-Row Cultivator

Dear Reviewer,

Thank you very much for your careful evaluation of our manuscript and for your positive and encouraging comments on the relevance, originality, and completeness of our work. We are grateful that you found the topic important for mechanical weeding and non-chemical crop management, and that you considered the methodology and conclusions to be sound. In the revised version, we have further clarified the main research question, strengthened the link between the results and the key agronomic implications, and improved the structure of the Discussion and Conclusion sections in response to all reviewers’ suggestions. Our detailed replies are provided below.

Best regards,

Haonan Yang, Liyan Wu*, Changyuan Zhai*

Comment # 1.1: What is the main question addressed by the research? The main question addressed by the research: How to develop an intelligent vision-based row-guided inter-row cultivator with electric lateral shifting?

Response to Comment 1.1 : Thank you very much for clearly summarizing the main question of our work. We fully agree with your understanding that the central question addressed by this study is how to develop an intelligent vision-based row-guided inter-row cultivator with electric lateral shifting. In the revised manuscript, we have slightly polished the Abstract and the final paragraph of the Introduction so that this objective is stated more explicitly for the reader.

The main question addressed by the research is how to design and validate an intelligent vision-based row-guided inter-row cultivator with electric lateral shifting that can:

1: use deep-learning-based row detection to obtain reliable crop-row position information in real time under field conditions;

2: compensate perception–communication–actuation delay through a control strategy so that the lateral-shift actuator can achieve centimeter-level row-guidance accuracy; and

3: ensure safe inter-row cultivation by quantitatively evaluating root-damage risk at different maize growth stages.

To answer this question, the study develops a complete system that integrates a YOLOv8s-based vision module, a servo electric-cylinder lateral-shift mechanism, a delay-compensation control model, and a unified-coordinate injury-arc evaluation method, and then verifies its performance through controlled-site offset tests and multi-stage field experiments.

Comment # 1.2: Do you consider the topic original or relevant to the field? Does it address a specific gap in the field? Please also explain why this is/ is not the case.

Response to Comment 1.2 : Thank you very much for your positive evaluation that the topic is original and relevant to the field of mechanical weeding and non-chemical crop management. We fully agree with your assessment. Our work also addresses a specific gap in the existing literature: most camera-guided inter-row cultivators focus on geometric row-tracking indicators (such as row-center deviation), but rarely quantify root-damage risk or determine an agronomically safe cultivation window. By combining a vision-based electric lateral-shift system with an injury-arc–based root-risk evaluation in a unified coordinate frame across different maize growth stages, the present study provides both an original system implementation and a quantitative framework for safe, precision inter-row cultivation.

The topic of this research is both original and highly relevant to the field of mechanical weeding. It is relevant because it targets vision-based, non-chemical inter-row cultivation of maize, which directly contributes to reducing herbicide use, potential CO₂ emissions, and negative impacts on field biodiversity. It is original in two main respects. First, the proposed system integrates a deep-learning–based row-detection module with an electric lateral-shift actuator and delay-compensation control to achieve centimeter-level row guidance under field conditions, whereas many existing systems are hydraulically actuated and focus primarily on geometric tracking. Second, the study introduces an injury-arc–based root-damage evaluation in a unified coordinate frame and applies it across different maize growth stages to identify a safe cultivation window. This explicitly links the kinematic performance of the row-guidance mechanism to agronomic indicators of root safety and weed-control coverage, addressing a gap in the literature where root-damage risk and timing of inter-row cultivation have seldom been quantified in a unified framework.

Comment # 1.3: What does it add to the subject area compared with other published material?

Response to Comment 1.3 : Thank you very much for pointing out that the manuscript provides a complete methodology from machine design to experimental execution and result interpretation for this original cultivator. We appreciate your positive assessment. In the revised version, we have kept this overall structure unchanged, and only made minor wording adjustments in the Materials and Methods and Results sections to make the progression—design of the vision-based electric lateral-shift system, planning of controlled-site and field experiments, and quantitative analysis of row-guidance accuracy and root-damage risk—clearer to the reader.

Compared with other published material, this work adds value to the subject area in three main ways. First, it presents a complete workflow for an original vision-based, electrically actuated row-guided inter-row cultivator, covering system design, experiment planning, experiment execution, and result analysis, rather than focusing on only one of these stages. Second, it integrates a deep-learning–based row-detection module with an electric lateral-shift mechanism and delay-compensation control, and evaluates the resulting row-guidance accuracy through both controlled-site offset tests and field trials. Third, it introduces a unified-coordinate, injury-arc–based method to quantify root-damage risk at different maize growth stages and links these risk levels to cultivation timing, thereby extending performance evaluation from purely geometric tracking indicators to agronomically meaningful safety metrics.

Comment # 1.4: What specific improvements should the authors consider regarding the methodology?

Response to Comment 1.4 : Thank you very much for your positive evaluation that the methodology presented in the manuscript is complete and does not require additional components. In line with your comment, we did not introduce any new methodological modules in the revised version. However, to improve clarity and reproducibility in response to other reviewers’ suggestions, we refined several methodological descriptions without changing the original experimental design or analysis pipeline. Specifically, we: (i) expanded the description of the image dataset and the training performance of the YOLOv8s-Seg model in Section 2.2.1 (data collection conditions, number of images, train/validation/test split, and final precision/recall/mAP); (ii) added a detailed explanation in Section 2.2 of how the perception–communication delay between the Jetson Orin NX and the C37 controller was experimentally calibrated; (iii) clarified in Section 2.3 that the root-zone diameters for the 2–3-, 4–5-, and 6–8-leaf stages were taken from previous studies under similar cultivation conditions [29,30], and explained how these diameters were used to construct the root-zone area and determine root injury; and (iv) introduced a short “Statistical Analysis” subsection (Section 2.4), stating how MAE, RMSE, SD and E95 were calculated and that one-way ANOVA on root-damage rate across growth stages was performed in Microsoft Excel with a significance level of p < 0.05. These refinements only clarify the existing methodology and do not alter the core methods or conclusions of the study.

Regarding possible improvements to the methodology, we agree with the reviewer that the current methodological framework is already complete and does not require additional components. Therefore, we did not add new methods in the revision. Instead, we focused on clarifying and documenting the existing procedures by elaborating the image dataset and YOLOv8s-Seg training setup, describing the experimental calibration of perception–communication delay, specifying the literature sources and usage of root-zone diameters, and explicitly stating the statistical analysis (including the one-way ANOVA on root-damage rate). These refinements strengthen the transparency and reproducibility of the methodology while keeping its structure unchanged.

Comment # 1.5: Are the conclusions consistent with the evidence and arguments presented, and do they address the main question posed? Please also explain why this is not the case.

Response to Comment 1.5 : Thank you very much for your careful reading and for your clear summary of the main conclusions. We are pleased that you find the conclusions consistent with the presented evidence and that they address the main question posed in the study. We fully agree with the key points you highlight, namely that the paper: (i) describes a vision-based row-guided inter-row cultivator for maize integrating a deep-learning–based row-line detection module, a lateral-shift servo electric-cylinder actuator, and a root-injury quantification method based on a unified coordinate frame and arc-length integration; and (ii) presents controlled-site offset tests (with MAE as a key indicator) and field experiments at different growth stages, where root-damage rate and coverage are analyzed at the 4–5- and 6–8-leaf stages. In the revised Discussion and Conclusion, we have slightly refined the wording to make the take-home messages you listed more explicit for the reader: that root-damage risk increases with plant growth, that larger lateral shifts imply higher mechanical loading and risk, that delay compensation of the actuators is essential for accurate steering, that the 4–5-leaf stage is the optimal weeding window for maize, and that after five leaves more elastic and shallower tools are required. These refinements do not change the results or interpretations, but help to more clearly convey the consistency between the evidence and the conclusions.

Yes, the conclusions are consistent with the evidence and arguments presented, and they address the main question of how to develop and validate an intelligent vision-based row-guided inter-row cultivator with electric lateral shifting. The study first introduces the system architecture, which combines a deep-learning–based row-line detection module, a servo electric-cylinder lateral-shift mechanism, and a root-injury quantification method based on a unified coordinate frame and arc-length integration. It then reports controlled-site offset experiments quantifying row-guidance accuracy using indicators such as MAE and E95, and field experiments at different maize growth stages quantifying root-damage rate and coverage. The Discussion and Conclusion link these results to several coherent findings: root-damage risk increases with plant growth; larger lateral shifts lead to higher mechanical loads and greater risk; actuator delay compensation is necessary to maintain centimeter-level guidance accuracy; the 4–5-leaf stage provides the optimal balance between root safety and weed-control coverage; and when maize plants have more than five leaves, shallower and more compliant tools are required. These findings are directly supported by the reported data and logically respond to the central research question.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

Dear Authors,

Your paper is mostly well written, but you still need to proofread it to make it more concise and enhance the flow and consistency. Some of the suggested corrections have been highlighted in the returned annotated manuscript.

Importantly, the comment made on the improvement of the Introduction section needs to be implemented to strengthen your work and highlight its significance.

Best wishes,

Dr. "Reviewer"

Comments for author File: Comments.pdf

Comments on the Quality of English Language

Proofread to remove redundancy and improve the manuscript’s conciseness and credibility.

Author Response

Design and Testing of a Vision-Based, Electrically Actuated Row-Guided Inter-Row Cultivator

Dear Reviewer,

On behalf of all the authors, I would like to express our sincere gratitude for your careful evaluation of our manuscript and for your thoughtful, constructive comments. We truly appreciate your positive assessment of the relevance and originality of our work on an intelligent vision-based row-guided inter-row cultivator for maize, as well as your recognition of the completeness of the methodology and the consistency of the conclusions with the presented evidence.

Your remarks on the importance of mechanical weeding and non-chemical crop management, and your clear summary of the main findings, have been very encouraging for us. In the revised version, we have carefully considered all of your suggestions: we clarified the main research question, improved the description of the experimental procedures and statistical analysis, and refined the Discussion and Conclusion to better highlight the key agronomic implications.

Thank you again for your time, expertise, and constructive feedback, which have helped us improve the quality and clarity of the manuscript.

Best regards,
Haonan Yang, Liyan Wu*, Changyuan Zhai*

Comment # 1.1: The idea of weeding is no longer restricted to mechanical means, which involves loosening the soil--there're now laser weeding and chemical (herbicide) technologies that do not necessarily impact the soil or "remove" (uproot) the weeds. It would be good to have that idea incorporated into your abstract.

Response to Comment 1.1 : Thank you very much for this insightful suggestion. We agree that modern weeding is no longer limited to mechanical soil cultivation, and that chemical herbicides and emerging technologies such as laser weeding should be acknowledged in the background. In the revised manuscript, we have modified the opening sentences of the Abstract to briefly introduce the broader spectrum of weeding technologies and then position maize mechanical inter-row cultivation as a key non-chemical option. The revised Abstract now reads:
“Modern weeding technologies range from chemical herbicides and emerging non-contact approaches such as laser weeding to conventional mechanical inter-row cultivation that loosens the soil and uproots weeds. Among these, maize inter-row cultivation and mechanical weeding remain important options for reducing herbicide use and improving the soil–crop environment.”

Lines:13-16

Modern weeding technologies include chemical weeding, non-contact methods such as laser weeding, and conventional mechanical inter-row cultivation characterized by soil loosening and weed uprooting. For maize, mechanical inter-row cultivation is key to cutting herbicide use and enhancing the soil–crop environment.

Comment # 1.2: Talking about precision, which is relative, provide an idea of the scale – compare this value with the crop row spacing; for example, as a percentage.

Response to Comment 1.2 : Thank you for this helpful remark. We agree that the lateral-adjustment capability should be expressed relative to the crop-row spacing to give readers a clearer sense of scale. In the revised Abstract, we have clarified that the ±15 cm value refers to the lateral travel range and explicitly related it to the 50 cm row spacing. The sentence now reads: “…issues control commands to a servo electric cylinder to achieve precise lateral adjustments within ±15 cm between rows, corresponding to 30% of the 50 cm row spacing.”

Lines:16-22

This study developed a vision-guided intelligent inter-row cultivator with electric lateral shifting: via maize plant recognition and crop row fitting, it computes real time lateral deviation between the implement and crop rows, and uses delay compensation to command a servo-electric cylinder for precise ±15 cm inter-row adjustments corresponding to 30% of the 50 cm row spacing.

Comment # 1.3: Use technical terminologies! V2 to V3 stage, V4-V5 stage, ...

Response to Comment 1.3 : Thank you for this useful suggestion. We agree that using standard phenological terminology makes the description of maize growth stages more precise. In the revised manuscript, we have replaced the informal expressions “2–3 leaf stage”, “4–5 leaf stage”, and “6–8 leaf stage” in the Abstract with the corresponding technical terms “V2–V3 stage”, “V4–V5 stage”, and “V6–V8 stage”. The same notation has been adopted consistently throughout the manuscript, while the first occurrence also keeps the leaf-count description in parentheses for clarity (e.g., “V4–V5 stage (4–5-leaf stage)”).

Lines:27-30

A root damage quantification method based on geometric overlap arc length was established, with rates rising with crop growth: 0.12% at V2 to V3 stage, 1.46% at V4 to V5 stage and 9.61% at V6 to V8 stage, making V4 to V5 stage the optimal operating window.

Comment # 1.4: You may also justify this based on any perceived benefits it could offer for mechanical harvesting, especially in comparison to other weeding technologies such as laser or chemical methods.

Response to Comment 1.4 : Thank you for this important suggestion. We agree that the advantages of the proposed system relative to other weeding technologies should be stated more clearly. In the revised Conclusion, we now explicitly contrast our approach with chemical and laser-based weeding methods. Specifically, we emphasize that, compared with chemical weeding, the proposed mechanical inter-row cultivation does not require herbicide applications, thereby avoiding issues related to residues, spray drift and the additional management burden associated with resistance development. Compared with laser weeding, which typically demands high power density at the tool and is limited in effective working width, the proposed tractor–implement system controls weeds across the full implement width and simultaneously performs shallow inter-row tillage in a single pass, facilitating integration with existing mechanized cultivation and harvesting operations. These additions clarify the distinctive advantages of the proposed system as a non-chemical option within integrated weed-management strategies.

Lines:30-35

Compared with chemical weeding, the system requires no herbicide application, avoiding issues related to residues, drift and resistance management. Compared with laser weeding which requires high tool power density and has limited effective width, the tractor–implement system enables full-width weeding and shallow inter-row tillage in one pass, facilitating integration with existing mechanized operations.

Comment # 1.5: May add "computer vision".

Response to Comment 1.5 : Thank you for this helpful suggestion. In the revised manuscript, we have added “computer vision” to the list of keywords so that the role of vision-based perception in the proposed system is more clearly reflected.

Lines:40

computer vision

Comment # 1.6: I would also want you to contrast mechanical weeding technology with other non-chemical weeding technologies, such as laser or electricity, thermal, flame, hot water, mulching, and seed timing (Early planting combined with transplanting--. Insert a detailed comparison table showcasing the merits and flaws of each technique, and provide a strong justification for why mechanical weeding remains a viable non-chemical contender despite its challenges. Line 34/35 would be a suitable location to start this.

Response to Comment 1.6 : Thank you very much for this insightful suggestion and for directing us to recent work on laser/electric, thermal–flame, and cultural non-chemical weed-management strategies. In the revised manuscript, we have expanded the Introduction to briefly contrast mechanical inter-row cultivation with other non-chemical weeding technologies, citing the recommended references separately as [4-6], and we have added a comparison table in the Appendix (Table A1) that summarizes the main principles, advantages, and limitations of mechanical weeding, laser/electric weeding, flame/thermal methods, mulching, and crop-timing measures. Based on this comparison, we now explicitly explain in the Introduction that, although mechanical inter-row cultivation faces challenges related to guidance accuracy and potential root injury, it remains a practical and competitive non-chemical option for large-scale row crops because it can be mounted on existing tractors, provides high field capacity at realistic field speeds, and directly integrates soil loosening with weed removal [4-6].

Lines:46-60

Recent progress in non-chemical weed management has produced several promising alternatives, including laser or electric weeding robots that thermally destroy individual weeds near the crop row [4], and flame or other thermal systems that kill seedlings by short, intense heat shocks [5]. Agronomic studies further emphasize cultural measures such as mulching, adjusted sowing dates and transplanting to reduce early weed pressure and herbicide demand [6]. However, these approaches typically face trade-offs in energy input, safety constraints, effective working width, or strong dependence on local climate and labor organization. By contrast, mechanically guided inter-row cultivation remains a robust non-chemical option that can be mounted on existing tractors, delivers high field capacity at practical speeds, and directly integrates soil loosening with weed removal. A concise comparison of these non-chemical weeding technologies is provided in Appendix A (Table A1). Therefore, despite its own challenges related to guidance accuracy and potential root injury, mechanical inter-row weeding continues to be a practical and competitive non-chemical contender in row crops.

Table A1. Comparison of mechanical inter-row cultivation with other non-chemical weeding technologies.

Technique (non-chemical)	Principle / typical targets	Main advantages (non-chemical benefits)	Main limitations / challenges
Mechanical inter-row cultivation (this study)	Steel sweeps or hoes cut, uproot or bury weeds between rows; shallow loosening of inter-row soil; typically combined with guidance system	No herbicides; high field capacity at practical travel speeds; can be mounted on standard tractors; directly integrates soil loosening with weed removal; robust under a wide range of field conditions	Risk of crop or root injury if guidance accuracy or timing is poor; less effective on in-row weeds; performance sensitive to soil moisture, clods and surface residues; requires careful matching of tools to crop stage
Laser / electric weeding	Vision system detects individual weeds close to the crop, followed by focused laser beam or electric discharge to thermally destroy the meristem	Very high spatial selectivity; minimal soil disturbance; suitable for intra-row weeds that are difficult to reach mechanically; no chemical residues	High capital and maintenance cost; narrow effective working width and relatively low forward speed; high energy demand; strict safety and eye-protection requirements; currently more suited to small areas or high-value crops
Flame / thermal weeding (incl. hot water / steam)	Propane flame, hot gases or hot water/steam rapidly heat seedlings and denature proteins, causing desiccation	Fully herbicide-free; can be effective against many small annual broadleaf weeds; applicable in organic systems; little mechanical disturbance of soil structure	High fuel consumption and associated CO₂ emissions; risk of fire and crop scorching; efficacy strongly affected by wind and rainfall; often requires repeated passes to control regrowth and perennial weeds; operator safety must be ensured
Mulching (organic, plastic or biodegradable)	Straw, compost, plastic film or biodegradable mulch forms a physical barrier at the soil surface, blocking light and physically hindering weed emergence	Strong suppression of new weed flushes; improves soil moisture retention and can buffer soil temperature; may substantially reduce herbicide demand in some systems	Cost and availability of mulch materials; plastic films require removal or recycling and may leave residues; mulches can interfere with mechanical operations and may increase certain pests (e.g., slugs or rodents) under humid conditions
Crop-timing strategies (early sowing, transplanting, stale seedbed, etc.)	Adjust sowing date or use transplants so that the crop emerges and closes canopy before the main weed flush; stale seedbed encourages weeds to germinate and be destroyed before planting	Very low direct input cost; can considerably reduce early-season weed pressure and herbicide use; compatible with both mechanical and chemical strategies; may contribute to more stable yields	Strongly climate- and site-dependent; planting too early can result in poor crop establishment; stale seedbeds require additional passes and machinery; does not directly remove later weed flushes and therefore usually needs to be combined with other control methods

Comment # 1.7: "target" or "control input"?? (FOV). And add it to the list of abbreviations.

Response to Comment 1.7 : Thank you for pointing out these wording issues. In the revised manuscript, we have replaced “target” with the more precise term “control input” in the description of Zhao et al. [16], so that the sentence now reads “converts vision-detected lateral deviation into a lateral control input for the weeding knife.” In Section 2.2, “field of view” is now written as “field of view (FOV)” at its first occurrence and FOV has been added to the list of abbreviations.

Lines:

Comment # 1.8: ... robust performance--rapid decay of ...

Response to Comment 1.8 : Thank you for this helpful remark regarding the description of the control objective and robust performance. In the revised manuscript, we have clarified the wording by explicitly linking robustness to the lateral-error dynamics, so that the sentence now reads: “During operation, the controller is tuned for robust performance under the constraint of minimal root-zone intrusion, characterized by rapid decay of the lateral deviation and stable tracking of the crop-row centerline, thereby achieving safe coverage and efficient maize inter-row cultivation, as shown in Figure 1.” We believe this revision makes the control objective and the meaning of “robust performance” much clearer.

Lines:151-154

During operation, the controller is tuned for robust performance under the constraint of minimal root-zone intrusion, characterized by rapid decay of the lateral deviation and stable tracking of the crop-row centerline, thereby achieving safe coverage and efficient maize inter-row cultivation, as shown in Figure 1.

Comment # 1.9: The 131 rated power is 1000 W, rated speed 3000 rpm, rated torque 3.18 N·m, and rated voltage 48 132 V, with a single-side travel of 15 cm.

Response to Comment 1.9:Thank you for pointing out the grammatical issue in the sentence describing the rated parameters of the servo motor. In the revised manuscript, we have rewritten this sentence for clarity and consistency: “The rated power, speed, torque, and voltage of the servo motor are 1000 W, 3000 rpm, 3.18 N·m, and 48 V, respectively, and the lateral actuator provides a single-side travel of 15 cm.” This revision removes the redundant “is,” unifies the sentence structure, and clearly presents all rated specifications.

Lines:171-173

The rated power, speed, torque, and voltage of the servo motor are 1000 W, 3000 rpm, 3.18 N·m, and 48 V, respectively, and the lateral actuator provides a single-side travel of 15 cm.

Comment # 2.0: Rephrase for smoothness and grammatical correctness.

Response to Comment 2.0: Thank you for pointing out that the original phrasing was not smooth and contained grammatical issues. In the revised manuscript (Section 2.2.1), we have rewritten this part for clarity and fluency. The relevant sentences now read: “The model trained in this study is a single-class detector targeting maize seedlings. After 100 training epochs, it achieved a precision (P), recall (R), and mean average precision (mAP) of 96.7%, 95.8%, and 96.2%, respectively. Crop row detection in the guidance system was implemented using a segmentation model based on YOLOv8s [23].” We believe this revision improves readability while accurately conveying the methodological details.

Lines:205-208

Comment # 2.1: constant-speed operation speed rephrase

Response to Comment 2.1: Thank you for pointing out that the phrase “constant-speed operation speed” was awkward and needed rephrasing. In the revised manuscript we have rewritten this sentence for clarity and fluency as:

Lines:259-263

Based on the constant forward speed of the implement and the fixed distance between the camera and the cultivator blade, the system calculates the trigger time for the servo-electric cylinder. In this way, when the lateral offset changes, the lateral-shift mechanism can reach the target position in time and with high accuracy.

Comment # 2.2: (MAE)--you still need to define the terms at their first instance even when you have a list of abbreviation at the end of the manuscript.

Response to Comment 2.2:Thank you for pointing out that statistical indices such as MAE should be defined at their first occurrence, even if they are later listed in the abbreviations section. In the revised manuscript, we have added the full names and abbreviations of all error metrics at their first appearance in Section 2.3, and we have checked the rest of the manuscript to ensure that MAE, RMSE, SD, E95, and maximum absolute error are consistently and explicitly defined at first use.

Lines:366-369

Based on this sequence, we report the mean signed deviation, mean absolute error (MAE), root mean square error (RMSE), standard deviation (SD), 95th percentile absolute error (E95), and maximum absolute error (Max |e|), and we also give the proportion of deviations falling within the ±15 cm and ±30 cm thresholds.

Comment # 2.3: Why is there an offset? Let the origin of each axis coincide.

Response to Comment 2.3: Thank you for pointing out the issue with the dual-axis plot in Figure 8. In the original version, the right-hand axis for average velocity started from a non-zero value, which visually created an offset between the two vertical axes. In the revised manuscript, we have redrawn Figure 8 so that both the total-time axis and the average-velocity axis start from zero and share the same origin on the x-axis. This removes the apparent offset and makes the time–displacement and velocity–displacement relationships clearer and more consistent with standard plotting practice.

Lines:474

Comment # 2.4: You could use more contrasting colors for this and the 0.51m/s line.

Response to Comment 2.4: Thank you for this helpful suggestion regarding figure legibility. In the revised manuscript, we have changed the color scheme of the row-guidance error curves so that the 0.51 m/s line and the other speed lines now use more contrasting colors, improving visual distinction between them.

Lines:499

Lines:507

Lines:516

Comment # 2.5: Move to after the paragraph starting at line 391. You may also increase the resolution (sharpness) of these graphs--Figures 9--11.

Response to Comment 2.5: Thank you for this practical layout suggestion. In the revised manuscript, we have moved Figures 9–11 to immediately follow the paragraph starting at line 391, so that each figure now appears directly after the corresponding description in the text. At the same time, we have replaced Figures 9–11 with higher-resolution versions to improve their sharpness and readability.

Comment # 2.6: Could you perform statistical (ANOVA) tests to verify whether there are indeed significant statistical differences between the speeds and offsets, or the observed differences are mere random effects.

Response to Comment 2.6:

Thank you very much for this important comment. In our original data processing, the root-damage rates at the 2–3-, 4–5- and 6–8-leaf stages were already evaluated using one-way analysis of variance (ANOVA), but this procedure and its outcome were not explicitly described in the manuscript. In the revised version, we have clarified this in Section 2.4 “Statistical Analysis”, where we state that the differences in mean root-damage rate among growth stages were tested by one-way ANOVA in Excel, with the significance level set at p < 0.05. In addition, the field experiment results in the subsection around Figure 14 and Table 3 have been updated to explicitly report the ANOVA findings: based on the data from three replicate passes, one-way ANOVA showed that growth stage had a highly significant effect on root-damage rate (p < 0.001); the mean root-damage rate at the 6–8-leaf stage was significantly higher than at the 2–3- and 4–5-leaf stages, whereas the difference between the 2–3- and 4–5-leaf stages was not statistically significant. These revisions provide a formal statistical basis for the reported values (0.12%, 1.46%, 9.61%) and strengthen the scientific validity of the conclusions.

Lines:442-459

Comment # 2.7: 10 cm is still centimeter level--change the wording. You could use "small", "negligible", or even "sub-centimeter".

Response to Comment 2.7: Thank you for pointing out that the phrase “centimeter-level row-guidance error” is ambiguous, since errors of several centimeters may still be described as “centimeter-level.” In the revised manuscript, we have replaced this wording with a more precise description of the actual guidance accuracy. The beginning of Section 3.3 now states that “even when the mean absolute row-guidance error is on the order of 1 cm, multiple local and continuous overlaps still occur in the middle and late stages,” which explicitly quantifies the error magnitude and avoids the misleading impression that the error is negligible.

Lines:613-615

Even when the mean absolute row-guidance error is on the order of 1 cm, multiple local and continuous overlaps still occur in the middle and late stages.

Comment # 2.8: State the implication of these factors-->non-generalizeability of your study.

Response to Comment 2.8: Thank you very much for pointing out that the limitations listed at the end of the manuscript should be accompanied by an explicit statement about their implications for the generalizability of our findings. In the revised version, we have added a clarifying sentence at the end of the limitation paragraph to state that, because of the restricted speed range, single-crop and relatively simple field conditions, and the lack of joint evaluation with yield and energy indicators, the current results should be regarded as evidence under the tested conditions rather than fully generalizable recommendations. We also note that additional field trials under a wider range of crops, terrains, and operating speeds are needed before the conclusions can be extended more broadly.

Lines:647-658

Comment # 2.9: Also provide details of your suggested future studies based on the outcome of this one.

Response to Comment 2.9: Thank you for the valuable comment. Based on the outcomes of this study, future research will focus on optimizing the force and attitude feedback mechanisms, refining the injury-risk–based closed-loop control strategy, and validating the system’s adaptability to other row crops, different cultivator configurations, and complex field scenarios (e.g., sloping fields, multiple soil types, and weed-dense plots) to further expand the technical application scope and stability under complex operating conditions.

Lines:680-683

Response to Comment 3.0:Thank you for your valuable comments; all format-related issues have been revised in accordance with your indications.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

This study involved the development of an intelligent, vision-based, row-guided inter-row cultivator with electric lateral shifting. The system relies on maize plant recognition and crop row fitting and computes the lateral deviation between the implement and the crop rows in real time. However, the following remarks should be considered before publishing

The abstract focuses on describing the technological system, which includes a vision-based cultivator, a servo-electric cylinder, plant recognition, and row-fitting algorithms. However, it provides insufficient information on how the study was conducted.

In 'Materials and Methods', the authors do not describe the statistical analysis procedures used in the study. It is essential to specify how the data was analyzed, including which statistical tests were applied, which software was used, the criteria for significance and any pre-processing steps (e.g. normalization and error calculation). Without this information, it is difficult to assess the validity of the results.

Several subsection headings in the Results section include the word 'Result' (e.g. 'Result of ...'). This is both unnecessary and inconsistent with standard scientific writing. Subsection titles should describe the content without repeating the word 'result'.

Reference citations are included in the Results section (lines 357–368). However, according to standard scientific writing conventions, the Results section should only present the findings of the current study and refer to figures, tables, and statistical outputs. References to external literature belong to the Discussion section.

In line 488, the manuscript refers to Figure 16, but based on the sequence and content, it should be Figure 14. Please revise the figure number accordingly.

Figures 12, 13, and 14 are unclear in their current form. The resolution is low; please improve the clarity and resolution.

The manuscript uses the heading 'Conclusions'. This section should be titled 'Conclusion' (singular).

I recommend that the authors briefly mention the study's limitations in the abstract and the conclusion. This will create a balanced view of the work's strengths.

Please ensure that all the abbreviations in the manuscript are defined when they first appear.

Author Response

Design and Testing of a Vision-Based, Electrically Actuated Row-Guided Inter-Row Cultivator

Dear Reviewer,

Best regards,

Haonan Yang, Liyan Wu*, Changyuan Zhai*

Comment # 1.1: The abstract focuses on describing the technological system, which includes a vision-based cultivator, a servo-electric cylinder, plant recognition, and row-fitting algorithms. However, it provides insufficient information on how the study was conducted.

Response to Comment 1.1 : Thank you for pointing out that the abstract focused too much on describing the technological system (vision-based cultivator, servo-electric cylinder, plant recognition, and row-fitting algorithms) and did not provide enough information on how the study was conducted. In the revised abstract, we kept only a concise description of the system and added a sentence that summarizes the experimental methodology. Specifically, we now state that a series of commanded lateral displacements from 1 to 20 cm were tested at travel speeds of 0.31, 0.42, and 0.51 m/s to characterize the time–displacement response of the servo-electric lateral-shift mechanism, and that field experiments at 0.51 m/s were conducted at three maize growth stages, with three 30 m passes per stage, to collect row-guidance error and root-injury data. These additions clarify how the dynamic tests and field trials were organized, so that the abstract reflects not only the system design but also the way the study was carried out.

Lines:22-27

To test the system’s dynamic response, 1–15 cm commanded lateral displacements were evaluated at 0.31, 0.42 and 0.51 m/s to characterize the time-displacement response of the servo-electric shift mechanism; field tests were conducted at 0.51 m/s with three 30 m passes per maize growth stage to collect row-guidance error and root-injury data. Field results show that at an initial offset of 5 cm, the mean absolute error is 0.76–1.03 cm, and at 15 cm, the -percentile error is 7.5 cm.

Comment # 1.2: In 'Materials and Methods', the authors do not describe the statistical analysis procedures used in the study. It is essential to specify how the data was analyzed, including which statistical tests were applied, which software was used, the criteria for significance and any pre-processing steps (e.g. normalization and error calculation). Without this information, it is difficult to assess the validity of the results

Response to Comment 1.2 : Thank you for your valuable suggestion to clarify the statistical analysis details. We have added a new subsection 2.4 Statistical Analysis in the revised manuscript to address this comment.

In this subsection, we explicitly specify the software (Microsoft Excel) used for calculating the statistical metrics (MAE, RMSE, SD, and E95) and detail the calculation logic of each metric. We also elaborate on the statistical method for root-damage risk analysis, including the calculation of root-damage rate, and confirm that one-way analysis of variance (ANOVA) was conducted to test the statistical significance of differences in root-damage rates among different maize growth stages, with the significance level set at p < 0.05.

This new subsection comprehensively supplements the statistical methodology of the study, ensuring the transparency and reproducibility of the data analysis process.

Lines:438-456

2.4 Statistical Analysis

Comment # 1.3: Several subsection headings in the Results section include the word 'Result' (e.g. 'Result of ...'). This is both unnecessary and inconsistent with standard scientific writing. Subsection titles should describe the content without repeating the word 'result'.

Response to Comment 1.3 : Thank you for your insightful comment regarding the subsection headings in the Results section. We fully agree that including the word "Result" in the subsection titles is unnecessary and inconsistent with standard scientific writing practices.Following your suggestion, we have revised all relevant subsection headings by removing the redundant "Result". The revised titles now directly describe the core content of each subsection, ensuring consistency with standard academic writing norms.

Lines:458,459,486,542

3.1 Controlled-Site Test

3.1.1 The Dynamic Response Test of the Row-Guidance Lateral-Shift Mechanism

3.1.2 The Row-Guidance Accuracy Test at Different Speeds

3.2 Field Trial

Comment # 1.4: Reference citations are included in the Results section (lines 357–368). However, according to standard scientific writing conventions, the Results section should only present the findings of the current study and refer to figures, tables, and statistical outputs. References to external literature belong to the Discussion section.

Response to Comment 1.4 : Thank you for your valuable comment regarding reference citations in the Results section. We fully agree with your point that, in line with standard scientific writing conventions, the Results section should only present the current study’s findings, figures, tables, and statistical outputs, while external literature citations belong to the Discussion section.Following your suggestion, we have removed all external reference citations (lines 357–368) from the Results section. The revised Results section now solely focuses on presenting the study’s own findings, relevant figures, tables, and statistical results, which is consistent with academic writing norms. Corresponding literature references have been appropriately adjusted to the Discussion section where necessary.

Lines:460-470

Under unchanged control parameters, 20 datasets with target displacements from 1 to 20 cm in 1 cm increments were analyzed. As shown in Figure 8, the total time for the servo-electric cylinder to complete one commanded displacement increases approximately proportionally with displacement. To quantify this trend, a least-squares linear regression was fit: each additional 1 cm of displacement increased the total time by 0.1426 s; moreover, as displacement approached zero, a fixed time of 0.025 s remained, attributable to start-up. The linear model achieved R² ≈ 0.997; the deviation between regression predictions and measurements was at most 0.0010 s (mean 0.00011 s), indicating that a linear relation is sufficient and robust for the time–displacement relationship. Taking the reciprocal of the time-per-centimeter slope yields a reference speed of about 7.014 cm/s, interpretable as the asymptotic overall average speed for sufficiently large displacements.

Comment # 1.5: In line 488, the manuscript refers to Figure 16, but based on the sequence and content, it should be Figure 14. Please revise the figure number accordingly

Response to Comment 1.5 : Thank you for your careful check and valuable comment. We have revised the figure number in line 488 as suggested: the incorrect reference to "Figure 16" has been corrected to "Figure 14" to align with the sequence and content of the figures in the manuscript.

Lines:589-590

Figure 14. Inter-row cultivation trajectory and root-zone distribution at the 6–8 leaf stage: (a) overview; (b) root-injury detail.

Comment # 1.6: Figures 12, 13, and 14 are unclear in their current form. The resolution is low; please improve the clarity and resolution.

Response to Comment 1.6 : Thank you for your careful observation and valuable comment. We have improved the clarity and resolution of Figures 12, 13, and 14 as suggested. The revised figures now feature sufficient resolution and clear details, fully meeting the requirements for academic presentation and publication.

Lines:553

Lines:563

Lines:587

Comment # 1.7: The manuscript uses the heading 'Conclusions'. This section should be titled 'Conclusion' (singular).

Response to Comment 1.7 : Thank you for your valuable comment. We have revised the section heading as suggested: the original plural form "Conclusions" has been corrected to the singular "Conclusion", which is consistent with standard academic writing conventions

Lines:656

4. Conclusion

Comment # 1.8: I recommend that the authors briefly mention the study's limitations in the abstract and the conclusion. This will create a balanced view of the work's strengths.

Response to Comment 1.8 : Thank you for your valuable suggestion. We agree that mentioning the study's limitations helps present a balanced view of the work. As recommended, we have briefly added the study's limitations in both the abstract and the conclusion sections, which enhances the objectivity and rigor of the manuscript.

Lines:34-37

These results, obtained at a single forward speed of 0.51 m/s in one field and implement configuration, still require validation under higher speeds and broader field conditions; within this scope they support improving the precision of maize mechanical inter-row cultivation.

Lines:675-680

However, these quantitative thresholds were derived from tests at a single forward speed of 0.51 m/s in one soil type and implement configuration, and should therefore be regarded as indicative rather than universally applicable. Future work will refine force and attitude feedback and injury-risk–based closed-loop control, and will validate the system’s adaptability to higher speeds, other row crops, and different cultivator geometries.

Comment # 1.9: Please ensure that all the abbreviations in the manuscript are defined when they first appear

Response to Comment 1.9 : Thank you for your careful comment. We fully agree that defining abbreviations at their first appearance is essential for improving the readability and clarity of the manuscript. Following your suggestion, we have systematically reviewed all abbreviations used in the manuscript (including technical terms, statistical indicators, and device-related abbreviations). For each abbreviation, we have added its corresponding full name and definition when it first appears in the text.

Author Response File: Author Response.pdf

Reviewer 5 Report

Comments and Suggestions for Authors

Please consider providing some more specific comments addressing the following points:

What is the main question addressed by the research?

The main question addressed by this research is how to develop a robust, high-accuracy, row-guided inter-row cultivation system that can operate effectively under complex field conditions (such as variable illumination, plant occlusion, and terrain undulation) while simultaneously minimizing root-zone injury to the crop.

The study specifically seeks to resolve the fundamental gap in current systems, which is the absence of:

Control strategies matched to actuator delay.
Lateral-magnitude constraints, per-cycle reachability checks, and suppression of large-offset uncertainty during correction.
Do you consider the topic original or relevant to the field? Does it address a specific gap in the field? Please also explain why this is/ is not the case.

The topic is highly relevant to the field of precision agriculture, particularly in the context of the global trend toward non-chemical weed control (AI-enabled mechanical weeding) to reduce environmental pollution from herbicides.

The paper addresses two specific gaps in current literature and technology:

Control System Gap: Existing vision-guided implements often treat the measured lateral offset as the commanded displacement, without compensating for the end-to-end latency (perception to blade). This leads to increased operational errors and risk of root injury, especially as travel speed increases. The study directly addresses this by building a coordinated control framework that incorporates delay compensation and bounded correction.
Quantification Gap: Most prior work focuses on alignment error and general outcomes, lacking a geometric quantification of root-zone spatial risk. The authors introduce a novel root-damage quantification method to systematically evaluate the risk

What does it add to the subject area compared with other published material?

The research makes several significant additions compared with other published material. It presents the design and testing of a new intelligent vision-based row-guided inter-row cultivator that uses an electrically actuated lateral shifting mechanism. It develops a control system that uses a forward-accounted delay compensation mechanism to execute lateral adjustments in advance, thereby mitigating the positional bias caused by the total system latency (perception, communication, and actuation). The control framework includes two-level saturation with hysteresis and travel buffering to provide bounded row-alignment correction, which is intended to prevent overshoot and reduce uncertainty associated with large offsets, a key deficiency in prevailing approaches. It introduces a method for quantifying root injury based on the geometric overlap arc length between the cultivation trajectory and the root system distribution at different leaf stages, providing a definitive metric for crop safety. The results indicate that the proposed system achieves greater operational stability and safer control at comparable accuracy levels to prior work.

What specific improvements should the authors consider regarding the methodology?

The authors clearly identify the following limitations in their discussion, which should be considered for future improvements to the methodology. The speed range for field trials was narrow, conducted only at the low speed of 0.51 m/s (1.83 km/h). The methodology would be strengthened by evaluating the effects of higher operating speeds on stability, accuracy, and root-injury risk. The tests were conducted exclusively in level maize fields. Future methodology should include trials in multi-crop scenarios and under complex terrain conditions to fully assess the system's robustness. The perception relied on vision and RTK. The authors suggest that adding pose or force feedback could help constrain attitude error and ground contact shocks, which would make the system more robust against terrain disturbances.

Are the conclusions consistent with the evidence and arguments presented and do they address the main question posed? Please also explain why this is/is not the case.

The conclusions are consistent with the evidence and arguments presented, and they address the main question posed. The argument for a high-accuracy, safe cultivator is supported by the results. The field trials showed that the mean absolute error (MAE) was low, ranging from 0.76 cm to 1.03 cm for a 5 cm offset, which validates the high accuracy claim and the effectiveness of the control strategy. Addressing the Main Question: The central problem of minimizing root injury is directly addressed by the finding that the 4–5 leaf stage is the optimal operating window, which exhibited a low root-damage rate of 1.46% (with the lowest at 0.12% for the 2-3 leaf stage). This evidence provides the "important theoretical basis and practical support" for precision inter-row cultivation that the paper aims to achieve

Are the references appropriate?

The references appear appropriate. The literature review covers essential background on the need for non-chemical weeding, prior work on vision-guided and intra-row weeding systems, and commercial implementations. The citations lay a clear foundation for the two major gaps in the study addresses (control delay and lack of root-injury quantification) and place the work within the current state-of-the-art.

Any additional comments on the tables and figures.
While the actual tables and figures are not available for direct inspection, their descriptions in the text suggest they are highly informative and essential for the paper's clarity:
- Clarity of Concepts: Figure 1 (Operating Principle) and Figure 4 (Control System Delay) are essential for making the core innovation, the integration of the mechanical system with the delay compensation control model (D=d-vt), conceptually clear to the reader.
- Mechanical Detail: Figure 2 (Structure Diagram) is necessary to explain the integration of the various components, such as the adjustable depth unit, tillage contour-following mechanism, and the lateral-shift unit with the servo-electric cylinder.
- Rigor of Results: The use of detailed quantitative metrics like MAE, RMSE, and E₉₅ in the tables ensures a rigorous evaluation of the row-guidance accuracy. The visual presentation of the cultivation trajectories in Figure 9, especially their near coincidence with the reference line at 5 cm offset, provides excellent visual evidence to support the accuracy claims.

Suggestion: The authors should ensure that all mechanical and control diagrams (e.g., Figures 1, 2, and 4) use large, legible labels and high-resolution quality so that the details of the innovative design are easily discernible.

Author Response

Design and Testing of a Vision-Based, Electrically Actuated Row-Guided Inter-Row Cultivator

Dear Reviewer,

Thank you sincerely for your meticulous evaluation of our manuscript and your positive, encouraging feedback on the relevance, originality, and comprehensiveness of our research. We greatly appreciate your recognition of the topic’s significance for mechanical weeding and non-chemical crop management, as well as your affirmation of the soundness of our methodology and conclusions. In response to all reviewers’ valuable suggestions, the revised version has further clarified the core research questions, strengthened the connection between the experimental results and key agronomic implications, and optimized the structure and logical flow of the Discussion and Conclusion sections. Detailed responses to each comment are provided below for your reference.

Best regards,

Haonan Yang, Liyan Wu*, Changyuan Zhai

Comment # 1.1: What is the main question addressed by the research?

Response to Comment 1.1: Thank you for formulating and summarizing the central research question so clearly. We fully agree that the core problem is how to develop a robust, high-accuracy row-guided inter-row cultivation system that can operate reliably under realistic field conditions while minimizing root-zone injury. In this work, we specifically target this question by (i) designing a vision-guided, electrically actuated lateral-shift mechanism with centimeter-level guidance accuracy and (ii) coupling it with a quantitative root-injury metric that links guidance performance to crop safety across different maize growth stages. We have clarified this focus in the Introduction and Abstract so that the main question is explicitly stated and directly connected to the experimental objectives and evaluation metrics.

Comment # 1.2: Do you consider the topic original or relevant to the field? Does it address a specific gap in the field? Please also explain why this is/ is not the case.

Response to Comment 1.2: We appreciate the reviewer’s positive assessment of the relevance and originality of the topic. We agree that the work is closely aligned with current developments in precision agriculture and non-chemical weed control, where AI-enabled mechanical weeding is becoming an important alternative to herbicide-based strategies. The manuscript is intended to address two specific gaps: (1) on the control side, the lack of coordinated strategies that explicitly account for end-to-end delay and bounded lateral correction, and (2) on the agronomic side, the absence of a geometric, root-zone–based risk quantification framework. In the revised manuscript, we have sharpened the statement of these gaps in the Introduction and more clearly positioned our contributions as bridging both the control-system and root-injury-assessment gaps identified by the reviewer.

Comment # 1.3: What does it add to the subject area compared with other published material?

Response to Comment 1.3: Thank you for your thoughtful summary of the contributions. Compared with existing work, the present study adds four main elements to the subject area. First, it reports the design and experimental validation of a vision-based, electrically actuated row-guided cultivator that achieves centimeter-level guidance and explicitly constrains the lateral correction range. Second, it implements a forward-accounted delay-compensation strategy that anticipates the total perception–communication–actuation latency, so that lateral adjustments are executed in advance rather than treating the measured offset as the instantaneous command. Third, it introduces a bounded-correction framework with saturation, hysteresis, and travel buffering to reduce overshoot and large-offset uncertainty. Fourth, it proposes a root-injury quantification method based on the geometric overlap arc length between the cultivation trajectory and growth-stage-dependent root-zone envelopes, thereby linking guidance performance to crop safety. We have emphasized these aspects more explicitly in the Discussion and Conclusion to highlight how the work extends current literature.

Comment # 1.4: What specific improvements should the authors consider regarding the methodology?

Response to Comment 1.4: We appreciate the reviewer’s constructive suggestions regarding methodological improvements. We agree that the current field experiments are limited to a relatively low forward speed (0.51 m/s) and to maize fields with moderate terrain variation, and that future studies should systematically investigate higher operating speeds, more complex terrain, and multi-crop scenarios. We also agree that augmenting vision and RTK with additional pose and/or force feedback would help constrain attitude errors and mitigate the effects of ground shocks on the implement. In response, we have expanded the limitations and future work sections to explicitly acknowledge the restricted speed range, field conditions, and crop scope of the present study, and we outline future methodological extensions involving higher-speed trials, multi-field validation, and the integration of force/pose sensing into a more comprehensive closed-loop control framework.

Comment # 1.5: Are the conclusions consistent with the evidence and arguments presented and do they address the main question posed? Please also explain why this is/is not the case.

Response to Comment 1.5: Thank you for confirming that the conclusions are consistent with the presented evidence and address the main research question. We agree that the low MAE (0.76–1.03 cm at a 5 cm offset) supports the claim of high guidance accuracy, and that the quantified root-injury rates across growth stages directly address the safety aspect of the system. In the revised Conclusion, we have strengthened the explicit link between the main question (accurate, safe, vision-guided inter-row cultivation) and the key findings: the centimeter-level guidance performance, the injury-arc–based risk metrics, and the identification of the 4–5 leaf stage as the preferred operating window under the tested conditions. At the same time, we now more clearly state the experimental limitations (single forward speed, single field and implement configuration), so that the conclusions are framed as well-supported within this scope rather than over-generalized.

Comment # 1.6: Are the references appropriate?

Response to Comment 1.6: We are grateful for the reviewer’s positive assessment of the reference selection. The intention was indeed to cover prior work on non-chemical weed management, vision-guided row and intra-row cultivation, and representative commercial systems, so as to highlight both the control-delay and root-injury-quantification gaps. In the revision process, we have kept this structure and have also added several recent references on non-chemical weeding technologies (including laser and other alternative approaches) to better situate our mechanical inter-row cultivation system within the broader context of emerging non-chemical weed control strategies.

Comment # 1.7: Any additional comments on the tables and figures.

Response to Comment 1.7: Thank you for the detailed and encouraging comments on the conceptual clarity and quantitative rigor of the tables and figures. We agree that the operating-principle and delay-model diagrams (Figures 1 and 4 in the manuscript) and the structural diagram (Figure 2) are central to understanding the integration of the mechanical system with the delay-compensation control model. Following your suggestion, we have checked all mechanical and control diagrams and trajectory plots to ensure that labels are sufficiently large, symbols are clearly distinguishable, and the exported resolution is suitable for print and online viewing. These adjustments are intended to make the innovative aspects of the design, control strategy, and experimental results more easily discernible to readers.

Author Response File: Author Response.pdf

Reviewer 6 Report

Comments and Suggestions for Authors

It is an interesting manuscript about the study and development of an intelligent vision-based row-guided inter-row cultivator with electric lateral shifting. It is a high technological design and development with a broad interest towards other similar machines and tasks. However, please consider the following comments and suggestions: i) the title maybe could be incorporate some final characteristic about the capability of the technology such as “remotely controlled” …” high precision” etc. depending of the case. ii) In the abstract; please mention materials and technology used, highlight the combination of components. Add further information related to trap the attention of the reader in the development and continue the reading. iii) in relation of the row guidance central system and mentioned equations that governs the movements, the system acts by the feedback response from the images generated within the path? Please consider to explain more about it. iv) Moreover; please consider to add images recorded for results discussed. v) about the camera; her position and cleaning required. Is there any particular procedure for optimal performance? The cleaning should be incorporated in the procedures? vi) about the discussion, is it possible to incorporate automatization to this system? Remotely control? vii) In the conclusions please check the definitions and meaning of errors; in general, the error in statistics is reported for a given magnitude as “the mean or average value ±the standard deviation”; by this manner the value could be involved within an interval of values. Please check that the interpretation of the measurements should be in similar manner done in this case.

Author Response

Design and Testing of a Vision-Based, Electrically Actuated Row-Guided Inter-Row Cultivator

Dear Reviewer,

Thank you again for your time, expertise, and constructive feedback, which have helped us improve the quality and clarity of the manuscript.

Best regards,
Haonan Yang, Liyan Wu*, Changyuan Zhai*

Comment # 1.1: Could the title incorporate some final characteristics about the capability of the technology (such as "remotely controlled", "high precision", etc.) depending on the case?

Response to Comment 1.1 : Thank you for your valuable suggestion on optimizing the manuscript title. We highly appreciate your insight that incorporating technological characteristics such as "high precision" could further highlight the system’s advantages, and we have carefully evaluated this proposal from both academic expression and research focus perspectives.

After thorough consideration, we have decided to retain the original title for three key reasons. First, a well-structured academic title should prioritize highlighting the core technical framework rather than stacking performance descriptors. The original title focuses on the fundamental innovation of this study: intelligent vision-based row guidance combined with electric lateral shifting. This focus allows readers to quickly identify the study’s technical orientation and core solution, which is more in line with the expression norms of engineering and agricultural machinery research titles. Second, the performance characteristics you mentioned such as high precision are not inherent attributes of the technical framework but outcomes verified through experiments. Including result-oriented terms in the title may blur the boundary between "technical design" and "experimental validation" and reduce the objectivity of the title. Third, we have systematically emphasized these key performance advantages throughout the manuscript to ensure readers do not miss them. For example, the abstract explicitly states centimeter-level guidance accuracy, the results section provides specific data on MAE and 95th-percentile error to validate high precision, and the discussion section contrasts with existing technologies to highlight the system’s stability and reliability. This arrangement follows the academic logic of "proposing technical solutions verifying performance advantages" and is more persuasive than simply adding descriptive words to the title.

We sincerely thank you for your thoughtful suggestion. We have confirmed that the core capabilities of the system are fully and rigorously presented in the main text, ensuring the manuscript’s academic rigor while maintaining the title’s conciseness and focus.

Comment # 1.2: 1. Please mention materials and technology used in the abstract and highlight the combination of components.

Add further information to attract the reader's attention to the development and encourage them to continue reading.

Response to Comment 1.2 : Thank you for your valuable and constructive suggestion on optimizing the abstract specifically regarding mentioning materials technologies component combinations and enhancing reader engagement. We have revised the abstract concisely to fully address your requirements explicitly noting core materials including Q235 low-carbon steel wear-resistant tool steel and aluminum alloy key technologies including YOLOv8s segmentation RTK fusion and delay compensation control and the perception-control-execution component combination including camera-RTK MCU servo-electric cylinder and adjustable shovels. We also incorporated specific experimental details including 1 to 15 cm displacements 0.31 0.42 and 0.51 meters per second speeds three 30-meter field passes and key results including 0.76 to 1.03 cm mean absolute error and 1.46 percent optimal root-injury rate to attract readers and highlight the system’s practical value aligning perfectly with your insights.

Lines:16-21

This study developed a vision-guided intelligent inter-row cultivator with electric lateral shifting—its frame fabricated from Q235 low-carbon structural steel and assembled mainly via bolted and pinned joints—that computes real-time lateral deviation between the implement and crop rows through maize plant recognition and crop row fitting, and uses delay compensation to command a servo-electric cylinder for precise ±15 cm inter-row adjustments corresponding to 30% of the 50 cm row spacing.

Lines:22-27

To test the system’s dynamic response, 1–15 cm commanded lateral displacements were evaluated at 0.31, 0.42 and 0.51 m/s to characterize the time-displacement response of the servo-electric shift mechanism; field tests were conducted at 0.51 m/s with three 30 m passes per maize growth stage to collect row-guidance error and root-injury data. Field results show that at an initial offset of 5 cm, the mean absolute error is 0.76–1.03 cm, and at 15 cm, the -percentile error is 7.5 cm

Comment # 1.3: Regarding the row guidance central system and the mentioned equations governing the movements: Does the system act based on the feedback response from the images generated within the path? Please explain this in more detail.

Response to Comment 1.3 : Thank you for your insightful question regarding the row guidance central system whether it operates based on the feedback response from path generated images as well as its working mechanism and governing equations. We confirm that the system relies entirely on image feedback for closed loop control and provide a detailed explanation as follows:

The system uses a camera to continuously capture real time images of maize plants and crop rows within the path which serve as the core input for feedback control. First the NVIDIA Orin NX platform combines deep learning algorithms to extract crop row information from the images calculate the geometric centerline of the rows and compare it with the camera’s field of view centerline to compute lateral deviation. Meanwhile the camera detects two maize plants in real time calculates the deviation between these plants and the current position of the lateral shift mechanism and transmits the offset to the C37 controller.

To ensure precise movement control the system integrates key parameters and delay compensation into its governing logic. Based on the implement’s constant forward speed v and the fixed distance d between the camera and the cultivator blade the controller calculates the trigger time for the servo electric cylinder. Additionally the system accounts for various delay factors including image processing data transmission and servo electric cylinder response time where lateral shift time is computed in real time and other delays are determined through experimental calibration. These delays and offset information are comprehensively calculated to trigger lateral shift actions at the precise moment correcting the lag between image recognition and execution control.

The entire process forms a continuous closed loop: after the servo electric cylinder adjusts the cultivator’s position the camera immediately captures new images to repeat the recognition calculation and adjustment cycle ensuring stable tracking of crop rows. The delay calculation formula referenced in the manuscript further quantifies this control logic.

We have supplemented these detailed mechanism explanations and associated equations in the revised System Working Principle section to enhance clarity fully addressing your inquiry

Comment # 1.4: Please consider adding images recorded to support the discussed results

Response to Comment 1.4 : Thank you for your valuable suggestion of adding recorded images to support the discussed results. We fully agree that visual evidence is important for verifying research findings, and we have carefully considered your proposal.

However, due to the satisfactory row-following accuracy of the cultivator system, the machine does not knock down or cause obvious damage to maize stalks during operation. The only potential seedling damage is limited to the maize root zone, which is not visually perceptible in real-shot field images—such images cannot intuitively reflect the actual seedling damage effect of the cultivator.

To address this issue and ensure objective and quantitative evaluation, this study adopts geometric quantitative diagrams generated from RTK coordinate data as the core evaluation basis. These diagrams can accurately characterize the spatial positional relationship between the cultivator and maize roots, enabling precise quantification of the damage degree. Compared with intuitive observation from real-shot images, this quantitative method is more reliable and rigorous for assessing the system’s row-following accuracy and seedling damage control effect.

Comment # 1.5: 1. What is the camera's position and the cleaning requirements for it?

Is there any specific procedure to ensure optimal performance?
Should cleaning be incorporated into the operating procedures

Response to Comment 1.5 : Thank you for your questions regarding the camera’s position, cleaning requirements, and related operating procedures. Below are detailed responses:

Camera’s position and cleaning requirements

The camera is mounted on the dedicated bracket on the left side of the implement in the forward direction (the camera’s installation position is shown in the two attached images). We have customized a shockproof, dustproof, and rainproof cover for the camera, which can effectively prevent dust, rainwater, and mechanical vibration from affecting the camera during operation. For cleaning, it is recommended to gently wipe the lens surface with a soft lint-free cloth to remove stains and ensure the clarity of image acquisition.

Specific procedure to ensure optimal performance

No specific operating procedure is required for the camera to maintain optimal performance. Its main function is to collect crop row images in real time; only need to ensure the lens is clean and the protective cover is intact, without additional parameter debugging or preprocessing steps.

Whether cleaning should be incorporated into operating procedures

The implement is equipped with an image quality recognition function. Before the implement starts working, it will automatically detect the quality of image acquisition. If the image is blurred due to lens contamination, the system will trigger an alarm prompt. In this case, operators are advised to clean the camera. Therefore, the cleaning step has been indirectly incorporated into the operating procedures through the system’s alarm mechanism, ensuring the device is in good condition before operation.

Comment # 1.6: In the discussion section: Is it possible to incorporate automation into this system? Can the system be remotely controlled?

Response to Comment 1.6 : Thank you for this helpful question regarding the level of automation and the possibility of remote control. At the current development stage, the proposed row-guidance system already performs automatic crop-row detection, lateral deviation computation, delay compensation, and servo-electric lateral shifting on board the implement controller, while the tractor driver only supervises the operation and adjusts forward speed locally. Remote control or teleoperation functions have not yet been implemented in this prototype. In the revised discussion, we have clarified this current automation level and added a short statement that future work will integrate wireless communication and farm-management interfaces to enable remote monitoring and higher levels of automation.

Lines:677-681

Future work will refine force and attitude feedback and injury-risk–based closed-loop control, and will validate the system’s adaptability to higher speeds, other row crops, and different cultivator geometries, and will investigate integrating wireless communication and remote monitoring to support higher levels of automation.

Comment # 1.7: 1. In the conclusions, please check the definitions and meanings of errors.

In general, error in statistics is reported as "the mean value ± standard deviation" for a given magnitude (to reflect the interval range of the value). Please ensure the interpretation of the measurements follows a similar format

Response to Comment 1.7 : Thank you for your valuable suggestion regarding the statistical expression of errors. We fully agree that presenting errors in the form of "mean ± standard deviation (SD)" is crucial for enhancing the rigor and reliability of research results, and we have carefully revised the relevant data presentation in the conclusions section in accordance with your advice.

Specific revisions are as follows:

For the controlled-site offset test results (core error-related indicators), we supplemented the statistical expression of "mean ± SD" as follows: Under low-speed and small-offset conditions, we added the standard deviation corresponding to the mean absolute error (MAE), i.e., "MAE was 0.76-1.03 cm (corresponding standard deviation (SD): 0.86–1.41 cm)"; we also added a specific operating condition example with "mean SD" expression: "for instance, at a 5 cm initial offset and 0.51 m/s, the lateral deviation was 0.76 0.86 cm (MAE SD, Table 2)". For the condition of 15 cm offset at 0.51 m/s, we supplemented the MAE with SD: "MAE = 1.94 ± 2.76 cm (MAE SD)".

Lines:662-667

For instance, at a 5 cm initial offset and 0.51 m/s, the lateral deviation was 0.76 0.86 cm. Across these conditions, the percentile absolute error was 1.5-3.1 cm, and the maximum absolute error was 3.5-4.0 cm; with a 15 cm offset at 0.51 m/s, the tail of the error distribution increased markedly(MAE = 1.94 2.76 cm; percentile absolute error ≈ 7.5 cm; maximum absolute error 9.0 cm.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

Dear Authors,

Thank you for your thoughtful revisions. You have significantly enhanced the quality of the manuscript. With the incorporation of the remaining suggested modifications, I am confident that the manuscript will be ready for publication.

Best wishes,

Dr. "Reviewer"

Comments for author File: Comments.pdf

Author Response

Design and Testing of a Vision-Based, Electrically Actuated Row-Guided Inter-Row Cultivator

Dear Reviewer,

Thank you again for your time, expertise, and constructive feedback, which have helped us improve the quality and clarity of the manuscript.

Best regards,
Haonan Yang, Liyan Wu*, Changyuan Zhai*

Comment # 1.1: This is probably not needed in the abstract. “its frame fabricated from Q235 low-carbon structural steel and assembled mainly via bolted and pinned joints”

Response to Comment 1.1 :

Thank you sincerely for your valuable comment and careful review of the abstract. We highly appreciate your insightful suggestion, which helps us optimize the content more effectively.

Considering that the core theme of this paper focuses on the design and development of the inter-row cultivator, the material selection (Q235 low-carbon structural steel) and connection method (bolted and pinned joints) of the frame are key technical details of the implement design. These details not only reflect the structural integrity and practicality of the designed cultivator but also help readers quickly grasp the core structural characteristics of the equipment, which is an important part of the design content highlighted in the paper.

Therefore, we hope to retain this part of the description in the abstract to ensure the completeness and accuracy of the key design information.

Thank you again for your constructive comment.

Comment # 1.2: Change field of view to FOV

Response to Comment 1.2 :

Thank you sincerely for your careful review and valuable suggestion. We have fully adopted your recommendation and revised "field of view" to its standard abbreviation "FOV" throughout the relevant content (e.g., abstract, main text). To comply with academic writing norms, "field of view (FOV)" is specified when mentioned for the first time, and the abbreviation "FOV" is used consistently thereafter to ensure format uniformity and readability.

Thank you again for your constructive comment, which helps improve the standardization of the manuscript.

Lines:144

FoV

Comment # 1.3: Lack of manufacturer details

Response to Comment 1.3 :

Thank you sincerely for your careful review and insightful comment. We fully acknowledge the lack of manufacturer details for the camera mentioned in the original sentence, and have supplemented the relevant information to ensure the traceability and completeness of the equipment description.

The revised sentence is as follows: "The images were captured using a Logitech C930c camera (Logitech Inc., Lausanne, Switzerland) at a resolution of 1920 × 1080."

We have added the core manufacturer details (company name and headquarters location) in parentheses, which complies with academic writing norms for equipment specification. Thank you again for your constructive suggestion, which helps improve the rigor and reproducibility of the manuscript

Lines:198-200

The images were captured using a Logitech C930c camera (Logitech Inc., Lausanne, Switzerland) at a resolution of 1920 × 1080.

Comment # 1.4: It should be two-class--"maize" or "no maize--weed".

Response to Comment 1.4 : Thank you sincerely for your insightful comment and careful review of the manuscript. We fully understand and highly agree with your suggestion that the detector should be designed as a two-class system, distinguishing between "maize" and "non-maize (weed)".

This adjustment aligns perfectly with the practical application scenario of our study—since the inter-row cultivator needs to accurately identify maize plants while distinguishing weeds to ensure precise operation. Your suggestion helps to optimize the experimental design and enhances the practicality and rigor of the research.

Thank you again for your constructive feedback, which contributes significantly to improving the quality of the manuscript.

Lines:205-207

The model trained in this study is a two-class detector designed to distinguish between two categories: "maize" and "non-maize (weed)".

Comment # 1.5: You Only Look Once, version 8, small (YOLOv8s)

Response to Comment 1.5 : Thank you sincerely for your careful review and valuable suggestion. We fully agree with your recommendation and have revised the expression of the model name throughout the manuscript.

Specifically, we have adjusted "YOLOv8s (You Only Look Once, version 8, small)" to "You Only Look Once, version 8, small (YOLOv8s)" as recommended. This revision complies with the academic writing norm of presenting the full name first followed by its abbreviation, ensuring clarity and readability for readers.

Thank you again for your constructive feedback, which contributes to improving the standardization and quality of the manuscript.

Lines:209-210

You Only Look Once, version 8, small (YOLOv8s)

Comment # 1.6: Add the version.

Response to Comment 1.6 :Thank you sincerely for your careful review and valuable suggestion. We fully understand that you recommend adding the version of Microsoft Excel, and we have supplemented the relevant information to ensure the rigor and reproducibility of the research.

Specifically, we have revised the relevant content to: "This error sequence was processed in Microsoft Excel 2021". This revision clarifies the software version used in the data processing, which aligns with academic writing norms.

Thank you again for your constructive feedback, which helps improve the completeness of the manuscript.

Lines:447-448

This error sequence was processed in Microsoft Excel (Version 2021).

Comment # 1.7: It will be good to include the ANOVA table showcasing the data.

Response to Comment 1.7:Thank you sincerely for your valuable suggestion in Comment 1.7. We fully agree that including the ANOVA table can enhance the clarity and rigor of the statistical results, so we have added Table 4 (titled "One-way ANOVA of root-damage rate among maize growth stages") in the manuscript.

This table is inserted alongside the textual description of the ANOVA results (specifically before the sentence introducing the ANOVA findings), and it presents detailed statistical metrics (including degrees of freedom, sum of squares, mean square, F-statistic, and p-value) related to the effect of growth stage on the root-damage rate.

Thank you again for your constructive feedback, which contributes to improving the completeness and readability of the manuscript.

Lines:613-614

Table 4. One-way ANOVA of root-damage rate among maize growth stages

Source of variation	degrees of freedom (df)	Sum of squares (ss)	Mean square (ms)	F-statistic (f)	p-value
Between stages	2	147.22	73.61	8.8 × 10²	0.001
Within stages	6	0.50	0.08	-	-
Total	8	147.72	-	-	-

Comment # 1.8: This comment requires replacing vague qualitative descriptions in the paper (such as "a narrow safety belt" and "within the centimeter range") with specific quantitative values to enhance academic rigor.

Response to Comment 1.8: Thank you for your valuable comment. To address the request for quantitative precision, we have revised the vague description "within the centimeter range" in the manuscript to "within less than 1 cm". This revision aligns with your suggestion of specifying a value below 1 cm, making the row-guidance error parameter more rigorous and verifiable. We appreciate your feedback for enhancing the accuracy of the manuscript’s technical descriptions

Lines:647

within less than 1 cm range

Comment # 1.9: Add FoV to the list.

Response to Comment 1.9: Thank you for your concise and helpful comment. We have added "FoV (Field of View)" to the specified list in the manuscript, ensuring the abbreviation is clearly defined and the list is comprehensive. We appreciate your attention to detail, which helps improve the completeness of the manuscript.

Lines:723

FoV Field Of View

Comment # 2.0: Add references justifying/supporting each of the points raised.

Response to Comment 2.0: Thank you for your constructive comment. To strengthen the academic rigor and credibility of the manuscript, we have meticulously added relevant references sentence by sentence for each point raised. All supplementary references are authoritative and closely aligned with the corresponding arguments, effectively justifying the rationality and validity of each viewpoint. We appreciate your feedback, which helps enhance the persuasive power and academic depth of the paper.

Lines:739

Technique (non-chemical)	Principle / typical targets	Main advantages (non-chemical benefits)	Main limitations / challenges
Mechanical inter-row cultivation (this study)	Steel sweeps or hoes cut, uproot or bury weeds between rows; shallow loosening of inter-row soil; typically combined with guidance system	No herbicides; high field capacity at practical travel speeds; can be mounted on standard tractors; directly integrates soil loosening with weed removal; robust under a wide range of field conditions [6]	Risk of crop or root injury if guidance accuracy or timing is poor; less effective on in-row weeds; performance sensitive to soil moisture, clods and surface residues; requires careful matching of tools to crop stage [6]
Laser / electric weeding	Vision system detects individual weeds close to the crop, followed by focused laser beam or electric discharge to thermally destroy the meristem [4]	Very high spatial selectivity; minimal soil disturbance; suitable for intra-row weeds that are difficult to reach mechanically; no chemical residues [4,6]	High capital and maintenance cost; narrow effective working width and relatively low forward speed; high energy demand; strict safety and eye-protection requirements; currently more suited to small areas or high-value crops [4,6]
Flame / thermal weeding (incl. hot water / steam)	Propane flame, hot gases or hot water/steam rapidly heat seedlings and denature proteins, causing desiccation [5]	Fully herbicide-free; can be effective against many small annual broadleaf weeds; applicable in organic systems; little mechanical disturbance of soil structure [5,6]	High fuel consumption and associated CO₂ emissions; risk of fire and crop scorching; efficacy strongly affected by wind and rainfall; often requires repeated passes to control regrowth and perennial weeds; operator safety must be ensured [4,6]
Mulching (organic, plastic or biodegradable)	Straw, compost, plastic film or biodegradable mulch forms a physical barrier at the soil surface, blocking light and physically hindering weed emergence [5,6]	Strong suppression of new weed flushes; improves soil moisture retention and can buffer soil temperature; may substantially reduce herbicide demand in some systems [5,6]	Cost and availability of mulch materials; plastic films require removal or recycling and may leave residues; mulches can interfere with mechanical operations and may increase certain pests (e.g., slugs or rodents) under humid conditions [5,6]
Crop-timing strategies (early sowing, transplanting, stale seedbed, etc.)	Adjust sowing date or use transplants so that the crop emerges and closes canopy before the main weed flush; stale seedbed encourages weeds to germinate and be destroyed before planting [6]	Very low direct input cost; can considerably reduce early-season weed pressure and herbicide use; compatible with both mechanical and chemical strategies; may contribute to more stable yields [6]	Strongly climate- and site-dependent; planting too early can result in poor crop establishment; stale seedbeds require additional passes and machinery; does not directly remove later weed flushes and therefore usually needs to be combined with other control methods [6]

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The authors have addressed all of my remarks, and I have no further comments.

Author Response

Dear Reviewer,

We are truly grateful to receive your feedback on our manuscript entitled “Design and Testing of a Vision-Based, Electrically Actuated Row-Guided Inter-Row Cultivator” (Manuscript ID: agronomy-4007451). Your confirmation that “all remarks have been addressed and no further comments are needed” is a great encouragement to our team.

From the initial review to this revision, your professional and detailed comments have been crucial in refining the scientific rigor of our work—whether it was optimizing the literature comparison of electric/hydraulic systems, clarifying root zone parameter sources, or supplementing the delay model sensitivity analysis, each suggestion has directly helped us improve the manuscript’s depth and clarity. We deeply appreciate the time and effort you have invested in evaluating our research; your insights have not only enhanced the quality of this paper but also provided valuable guidance for our future work in intelligent agricultural machinery.

We will continue to maintain the rigor of our research and look forward to the further progress of the manuscript. Thank you again for your valuable support and professional guidance.

Sincerely,

Changyuan Zhai

Corresponding Author

Intelligent Equipment Research Center, Beijing Academy of Agriculture and Forestry Sciences

Beijing 100097, China

Author Response File: Author Response.docx