Field Rice Growth Monitoring and Fertilization Management Based on UAV Spectral and Deep Image Feature Fusion
and Chenglong Huang 1,2,3,4,*Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsReview report.
Title
I recommend that the authors rewrite the title of the manuscript to make it more concise. For example: Management strategies based on fusion of spectral features and UAV-3 image used in monitoring rice growth and fertilization in the field.
Abstract
I ask the authors to rewrite some parts of the abstract. For example: in line 22, the development stages can be described in a technical way. After all, what are the crop development stages that the authors are referring to? The rice development cycle is divided into three phases, comprising: vegetative phase (from emergence to panicle differentiation), reproductive phase (from panicle differentiation to anthesis) and grain filling phase (from anthesis to physiological maturity) as cited by (GAO et al., 1992; INFELD et al., 1998).
Introduction
I liked the review of the state of the art (introduction) but I noticed the lack of information about the challenge of uniformizing crop fertilization during the preparation of rice cultivation. This technique can provide a lot of information about the nutritional status of plants and their development. However, it is necessary to understand that the lack of uniformity of the soil can generate zones of undesirable vegetative development and, depending on the stage of development of the crop, the solutions may become unfeasible. So, the technique is undoubtedly a very powerful strategy, but some empirical concepts need to be analyzed together. This is a suggestion I make to complete the meaning of the text in the introduction.
Materials and methods
In Table 1, the authors describe the NPK concentrations administered in each experiment. However, we know that all good experimental planning begins with a thorough review of the state of the art. The work carried out here is remarkable and of great importance for the development of systems for detecting nutrient deficiencies in crops. Therefore, I noticed the lack of information on soil analysis, as well as the correct description of the fertilizers used in the different stages of the experimental execution. The authors name them pure N, pure P and pure K. Fertilizers usually have specific formulas in their contents, and in my opinion Table 1 needs to be better described. Note that if errors occur in the experimental basis, all the meticulous work carried out will be compromised.
Results
The results obtained by the authors are encouraging and commendable. However, the research is focused on nutrient deficiencies in rice crops. We know that macronutrient deficiencies are not always the cause of plant development, and that the decline in soil fertility depends on many factors other than the fertilizers used in field management. My view is that false positives should be taken into account when analyzing the data obtained, since the deep neural network depends on coherent information to generate data that represents the reality of the crop in the field. Therefore, I would like to congratulate the authors for the great effort in focusing on the development of this line of research, but they should take into account the set of factors that can induce errors in the nutrient deficiency detection system.
Author Response
Comment 1：
I recommend that the authors rewrite the title of the manuscript to make it more concise.
Reply 1：
Thank you for your valuable and thoughtful comments. We have changed the title to "Field Rice Growth Monitoring and Fertilization Management Based on UAV Spectral and Deep Image Feature Fusion," which is more concise while retaining the key elements we aim to highlight.
Comment 2：
I ask the authors to rewrite some parts of the abstract. For example: in line 22, the development stages can be described in a technical way.
Reply 2：
Thank you for your valuable and thoughtful comments. We have briefly described the basic experimental design, image acquisition methods, and the specific growth stage of rice when the images were collected (as you mentioned) in the abstract. The revised content is as follows, and you may also refer to lines 20-22 in the manuscript.
“The deep image features achieved nutrition classification accuracies of 88.78% and 84.56% for rice spikelet-protection fertilizer application stage (S1) and bud-promoting fertilizer application stage (S2).”
Comment 3：
I liked the review of the state of the art (introduction) but I noticed the lack of information about the challenge of uniformizing crop fertilization during the preparation of rice cultivation.
Reply 3：
Thank you for your valuable and thoughtful comments. In response to your suggestion, we have reviewed relevant literature and recognized that this is indeed an important aspect requiring attention. Accordingly, we have added related discussions in the Introduction section. The specific modifications are as follows, and you may also refer to lines 84-87 in the manuscript.
“However, field fertilization practices do not exhibit ideal uniform distribution [38], but rather demonstrate variability depending on application methods and crop growth stages [39]. And excessive or indiscriminate fertilizer application can adversely affect land productivity [40, 41].”
References:
[38] Wang, W.H.; Cai, L.L.; Peng, P.Y.; Gong, Y.D.; Yang, X.Q. Soil Sampling Spacing Based on Precision Agriculture Variable Rate Fertilization of Pomegranate Orchard. Commun. Soil Sci. Plan. 2021, 20, 2445–2461.
[39] Siqueira, R.; Mandal, D.; Longchamps, L.; Khosla, R. Assessing Nitrogen Variability at Early Stages of Maize Using Mobile Fluorescence Sensing. Remote Sens. 2022, 14, 5077. https://doi.org/10.3390/rs14205077
[40] Wang, H.; Xu, J.; Chen, B.; Li, Y.; Li, S.; Liang, H.; Jiang, Q.; He, Y.; Xi, W. Performance of an Automatic Variable-Rate Fertili-zation System Subject to Different Initial Field Water Conditions and Fertilizer Doses in Paddy Fields. Agronomy 2023, 13, 1629. https://doi.org/10.3390/agronomy13061629
[41] Zhou, P.; Ou, Y.; Yang, W.; Gu, Y.; Kong, Y.; Zhu, Y.; Jin, C.; Hao, S. Variable-Rate Fertilization for Summer Maize Using Combined Proximal Sensing Technology and the Nitrogen Balance Principle. Agriculture 2024, 14, 1180. https://doi.org/10.3390/agriculture14071180
Comment 4：
In Table 1, the authors describe the NPK concentrations administered in each experiment. However, we know that all good experimental planning begins with a thorough review of the state of the art. The work carried out here is remarkable and of great importance for the development of systems for detecting nutrient deficiencies in crops. Therefore, I noticed the lack of information on soil analysis, as well as the correct description of the fertilizers used in the different stages of the experimental execution. The authors name them pure N, pure P and pure K. Fertilizers usually have specific formulas in their contents, and in my opinion Table 1 needs to be better described. Note that if errors occur in the experimental basis, all the meticulous work carried out will be compromised.
Reply 4：
Thank you for your valuable and thoughtful comments. In accordance with your suggestions, we have revised the specific fertilization measures mentioned in Table 1 and Table 6. Please refer to lines 139 and 555 in the manuscript for details. We have clarified the types of fertilizers used at different growth stages, primarily Compound fertilizer, Urea, and Potassium chloride, and also corrected the unit conversions for fertilization amounts.
Regarding the concern about "the lack of information on soil analysis," we did not conduct soil testing prior to the experiment because the control and treatment groups were established at the same location, leading us to overlook potential differences in initial soil fertility. However, your comment is highly valuable, and we will ensure more rigorous experimental design procedures, including soil analysis, in future field studies.
Comment 5：
The results obtained by the authors are encouraging and commendable. However, the research is focused on nutrient deficiencies in rice crops. We know that macronutrient deficiencies are not always the cause of plant development, and that the decline in soil fertility depends on many factors other than the fertilizers used in field management. My view is that false positives should be taken into account when analyzing the data obtained, since the deep neural network depends on coherent information to generate data that represents the reality of the crop in the field. Therefore, I would like to congratulate the authors for the great effort in focusing on the development of this line of research, but they should take into account the set of factors that can induce errors in the nutrient deficiency detection system.
Reply 5：
Thank you for your valuable and thoughtful comments. You raised an important point regarding the complexity of field conditions, where crop growth results from multiple interacting factors, and the inherent limitations of our classification model. We acknowledge these as valid shortcomings of the current study. Accordingly, we have included a critical analysis of these limitations in the Conclusions section to inform and guide our future research directions.
Reviewer 2 Report
Comments and Suggestions for Authors2.Materials and Methods
1) Line 171, in table 2, identify what the acronyms mean in a new column and in another column identify the author of each formula.
3.Results and Discussion
2) In Figure 9, identify what the acronyms mean
4. Conclusions
3) In the conclusion section, there is a missing perspective related to future research work
4) Please write about the limitations of this work in details in conclusion section?
Author Response
Comment 1：
Line 171, in table 2, identify what the acronyms mean in a new column and in another column identify the author of each formula.
Reply 1：
Thank you for your valuable and thoughtful comments. We have supplemented the content as suggested, and the detailed additions can be found in Line 177 and Table 2 of the manuscript.
Comment 2：
In Figure 9, identify what the acronyms mean.
Reply 2：
Thank you for your valuable and thoughtful comments. We have supplemented the content as suggested by adding a Note below Figure 9 for clarification. The specific additions are as follows, and can also be found in Line 521 of the manuscript.
“Note: R indicates feature combination of RGB-VIs, S indicates feature combination of Spec-tral-VIs, and F indicates deep image features.”
Comment 3：
In the conclusion section, there is a missing perspective related to future research work. Please write about the limitations of this work in details in conclusion section?
Reply 3：
Thank you for your valuable and thoughtful comments. In response to your suggestions, we have expanded the Conclusion section to provide a more comprehensive discussion of the current study's limitations and future research directions.
Reviewer 3 Report
Comments and Suggestions for AuthorsAbstract: The processes to be developed in the materials and methods section should be indicated in order to understand the work methodology to be developed.
Introduction: Consider the contribution made by each of the reviewed articles. The intention is to indicate the procedures and trends, indicated as the state of the art, that should be considered in order to propose our research. Each of the articles in the reference should be considered.
Materials and methods: It is recommended to indicate a map of how the proposal will be developed; this can be aided by a flowchart, block diagram, among others. It should indicate how the dataset construction process and the feature extraction processes are carried out. For model evaluation, the sensitivity and specificity of the model can be used as an aid. Each figure and table should be explained in order to analyze their contribution to understanding the methodology.
Results: It is recommended to use sensitivity and specificity to evaluate the models. Each graph should indicate its interpretation and how it helps define the proposed model.
Conclusions: Consider the benefits of the proposed methodology and how it helps or contributes to improving corn planting and harvesting.
References: Consider only academic articles. Check how articles are referenced on the journal's website. It is recommended to use references based on publications from the last three years.
Comments on the Quality of English LanguageIn accordance with English
Author Response
Comment 1：
The processes to be developed in the materials and methods section should be indicated in order to understand the work methodology to be developed.
Reply 1：
Thank you for your valuable and thoughtful comments. We have briefly outlined the Schematic diagram of the overall experimental design and technical approach as shown in figure 1 in the materials and methods section.
Comment 2：
Consider the contribution made by each of the reviewed articles. The intention is to indicate the procedures and trends, indicated as the state of the art, that should be considered in order to propose our research. Each of the articles in the reference should be considered.
Reply 2：
Thank you for your valuable and thoughtful comments. In the Introduction section, we have revised the content according to your suggestions to more clearly indicate the contributions of each reference in different fields. For example, we have refined the original statement: “Over the past decade, UAVs have been widely adopted in precision agriculture [7], facilitating applications such as pest detection [8], crop health assessment [9], yield prediction [10] and nutrient management [11, 12].” This modification provides more precise citations that better reflect each study's specific contribution. The revised version can be found in the manuscript.
Comment 3：
It is recommended to indicate a map of how the proposal will be developed; this can be aided by a flowchart, block diagram, among others. It should indicate how the dataset construction process and the feature extraction processes are carried out.
Reply 3：
Thank you for your valuable and thoughtful comments.
Regarding the overall technical approach of this study, detailed explanations are provided at the beginning of Chapter 2, accompanied by a supporting figure. Please refer to Figure 1 in the manuscript for specific content.
Concerning the dataset construction, Section 2.2 provides comprehensive descriptions covering the entire process from data collection and screening to dataset establishment.
For the feature fusion process, we acknowledge that the previous description might have been insufficient. Following your suggestion, we have enhanced the explanation in Section 2.4 with additional details about feature fusion. The specific additions are as follows, and may also be found in lines 294-303 of the manuscript.
“Using the trained Modified-VGG16 model, the forward propagation function was employed to extract the output of the final fully connected layer as deep image features (represented as a set of one-dimensional vectors). Subsequently, VI features were fused with the deep image features into a unified set of one-dimensional feature vectors, followed by normalization to achieve feature fusion.”
Comment 4：
It is recommended to use sensitivity and specificity to evaluate the models. Each graph should indicate its interpretation and how it helps define the proposed model.
Reply 4：
Thank you for your valuable and thoughtful comments. Following your suggestions, in addition to classification accuracy, we have introduced recall rate and F1-score for comprehensive evaluation of the classification models. The original formulas are presented in lines 323-330 of the manuscript. Furthermore, we have added F1-score as an evaluation metric for different classification models in Table 5 (line 466), along with corresponding analysis and discussion. The specific modifications are as follows, and you may also refer to the content in lines 493-500 of the manuscript.
“The F1-score analysis demonstrates that the RF model consistently outperforms SVM and XGBoost across most feature combinations, achieving the highest scores (F1-score: 0.808–0.976). Although XGBoost exhibits comparable performance (F1-score: 0.732–0.974), it underperforms with simpler features and shows greater performance variability. SVM achieves the least favorable results overall (F1-score: 0.708–0.949), particularly demonstrating significant limitations when handling complex feature sets. Consequently, RF exhibits superior robustness in managing diverse feature combinations, establishing it as the optimal classification model for this study.”
In response to your observation, we have added an analysis of the confusion matrix results in Figure 8 and their implications for evaluating the deep learning model's performance. The specific discussion is as follows, and may also be found in lines 439-447 of the manuscript.
“However, since VI features and deep image features represent significantly different crop physiological mechanisms and belong to distinct feature types, this is evidenced in Fig. 8, where the deep learning approach demonstrates superior perfor-mance in both low-fertilization (N1) and high-fertilization (N4) zones. This enhanced performance can be attributed to the more pronounced manifestation of color features under extreme fertilization conditions. In contrast, greater classification errors were observed in medium-fertilization zones (N2 and N3), particularly in the N3 experi-mental area. Therefore, this study proposes that combining these two feature types could theoretically achieve improved classification accuracy.”
Comment 5：
Consider the benefits of the proposed methodology and how it helps or contributes to improving corn planting and harvesting.
Reply 5：
Thank you for your valuable and thoughtful comments. In response to your suggestions, we have revised the Conclusion section to provide clearer and more detailed descriptions of the practical utility of the proposed method. The specific discussion is as follows, and may also be found in lines 622-626 of the manuscript. Additionally, we have further elaborated on future research directions.
“Based on the optimal classification model, the actual nutrient deficiency status in the field trial area was evaluated. Fertilizer supplementation was applied to regions with insufficient soil fertility, resulting in the generation of a site-specific fertilization prescription map suitable for rice growth. Implementation according to this prescription map effectively optimized crop growth performance.”
