Dynamic Measurement of Portos Tomato Seedling Growth Using the Kinect 2.0 Sensor
, Alfonso Ramírez-Pedraza 2,3,*, Francisco-Javier Ornelas-Rodríguez 1, Diana-Margarita Cordova-Esparza 4 and Erick-Alejandro González-Barbosa 5Round 1
Reviewer 1 Report
This paper proposes an automated ML-based approach for measuring the growth of portos tomato seedlings. Fundamental to the process being proposed is the Kinect 2.0 sensor, which has the potential to replace the expert-driven manual approach that is usually adopted at present. The paper needs significant restructuring as many important points raised by the authors are being lost in a maze of textual descriptions. Section 1 reads like a related research section. I suggest writing a brief new introductory section that clearly explains the motivation for this research. The current section 1 (introduction) should become a related research section. The novelty and contribution of this research should be articulated clearly at the very end. Section three should address technologies (3D segmentation etc.) and the methodology. At present, these details are conflated in sections 1, 2, 3 and 4, which makes it very confusing for the reader. Some points in the results (Section 5) are more suited to the methodology. The results sections should include the results and be followed by a discussion. The current section 6.1 is very good and should conclude the discussion. Finally, the conclusion section should succinctly summarise the research and propose some further avenues of research. Overall, the paper could be shortened and thus more focused. A thorough language review is needed.
Author Response
The study shown in this article proposes a new methodology for monitoring the growth and development of porto tomato seedlings. The system can predict crop production from the 15th day. Trained personnel divide the seedlings into three stages of growth and failure. Based on experience, they show that the 15th day after planting is the ideal day for nutrient application and irrigation decisions. An economical, easily transportable system is proposed without modifying greenhouse conditions. The average greenhouse error for yield estimation is plus minus $\pm2\%$. In our study, we reduced this error by up to $1\%$. We evaluate optimising and comparing various methods at each critical stage, segmentation and grading, to determine which is most efficient under the given conditions. The system will monitor $100\%$ of seedlings without fatigue in the future. Finally, uncertainties are not expressed because they are considered intrinsically in the decision parameters.
Reviewer 1
This paper proposes an automated ML-based approach for measuring the growth of portos tomato seedlings. Fundamental to the process being proposed is the Kinect 2.0 sensor, which has the potential to replace the expert-driven manual approach that is usually adopted at present.
The paper needs significant restructuring as many important points raised by the authors are being lost in a maze of textual descriptions.
1.- Section 1 reads like a related research section.
I suggest writing a brief new introductory section that clearly explains the motivation for this research. The current section 1 (introduction) should become a related research section. The novelty and contribution of this research should be articulated clearly at the very end.
A: We add a new paragraph in lines 24-30 describing the motivation for the work. Section 1 was placed as related research. Section two shows the state-of-the-art requested by reviewer 2, in comment 2. The results demonstrate the performance of the predictive system, the discussion and conclusions explain the novelty and contribution of the work.
2.- Section three should address technologies (3D segmentation etc.) and the methodology. At present, these details are conflated in sections 1, 2, 3 and 4, which makes it very confusing for the reader. Some points in the results (Section 5) are more suited to the methodology.
A: We move the explanation of Figures 3, 4, 5 \and{and} Table 1 on section 3, Lines 180-183, 195-203 and 204-223.
3.- The results sections should include the results and be followed by a discussion.
A: Comment applied.
4.- The current section 6.1 is very good and should conclude the discussion.
A: Thanks for your comment. We include the discussion section on lines lines 558-582.
5.- Finally, the conclusion section should succinctly summarise the research and propose some further avenues of research.
A: We moved two paragraphs of the conclusion, and both we added to the discussion, lines 536-542 and 551-556. And the conclusion section succinctly summarises and propose some further avenues of research, lines 586-601.
6.- Overall, the paper could be shortened and thus more focused.
A: We removed lines 293, 312, 330, 347, 353. The authors consider it is necessary to leave the equations as a foundation for the methodology. In addition, the rest of the information in the introduction, related works as they complement and strengthen the study presented here. However, if you consider irrelevant information, you could let us know and delete it. On the other hand, reviewer 2 requested modifications that enhanced the research study presented here.
7.- A thorough language review is needed.
A: Thanks for your comment. English writing and grammar improved.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This paper explains designing and implementation of a dynamic measurement system using Kinect 2.0 services. The author utilizes the GMM, Mean Shift, DTC to achieve automatic segmenting and classifying portos tomato seedlings. This study also illustrates the crop production accuracy of the growth stage obtained by each classifier compared to the crop production accuracy of experts. Experimental studies like this one, which also involve construction and experiments, require a significant effort and investment of time and resources. Overall the paper is logically structured and the work is presented in a reasonable manner. The conditions for dynamic measurement system evaluation were appropriate designed. However, I would like to see a more in depth description of the novelty of this work, and how it relates to previous work in the field of phenotyping system with crop growth mapping and measurement services. The following are some suggestions.
Introductions describes the actuality of the technological solution, but does not reveal the novelty of the question or scientific relevance.
Improve the literature review on segmentation, feature extraction, and cassifier applied to dynamic measurement systems specially the greenhouse environment and greenhouse crops growth. This is required in order to justify the proposed approach. Extend the review to imaging systems applied to model vegetables crop growth.
Please include a nomenclature defining all variables, parameters and abbreviations before the introduction.
Methodology section contain mathematical description of the technological solution (segmentation, feature extraction, and classifying) of production of portos tomato services, but the performance should be quantitatively evaluated so that it can be compared with other works. My point is that any dynamic measurement model of phenotyping system must take into account the existence of those several time-constants. As a matter of fact, one attractive feature of the proposed dynamic measurement architecture is the combination of several sub-model (segmentation, extraction and classification) and authors should connect their design with the time-scales found in the greenhouse phenotyping and measurement system. My suggestion is for authors to use short sampling or scan times (Kinect v2) to check the performance and behavior of their dynamic measurement systems.
Please don't start the paragraph with a sentence that just says the Figure shows specific information - you should use the main text to explain what you found in your research, then cite a Fig that provides explanatory details.
Dynamic measurement system with Kinect v2 services itself is presented in section 3, 4, amd 5 in detail, however, it does not solve or analyze any of scientific uncertainties or authors did not disclose that in the manuscript.
The way you present Result is a bit strange. You should present all measurements and analysis methods of the experiment first (GMM, DTC, etc.) and then the experimental set-up. Here you miss the analysis of the methodology to measure the surface of the leaves (stem of height) in points, the number of the leaves and so on and how you determined the optimum values presented in Fig. 9.
System implementation. Authors ought to provide some details of how the proposed dynamic measurement systems was programmed in a computer (GUI) system.
as Discussion is too short, it should be rewritten in accordance to results of statistical analyses (especially in terms of computation load and time with different dynamic measurement system) and en-reached about several recent works. It should be noted that you need to use references to explain your results and discuss them when you are trying to explain something found and not presented in previous work. The content of Section 6.1 is nothing new and does not show some scientific uncertainties and discoveries.
The author does not describe in detail how to obtain the comparative data in Tables 3 and 4, including the indicators that define the performance evaluation (crop production accuray and model accuracy)?
Others
- The authors do not describe in detail how the Kinect platform works in the greenhouse, including how the hardware device performs large-scale scanning work, how it moves, and what conditions are required?
- Although the author has described the theory required in the proposed method, the implementation details of the six stages and the names of the processing software or statistical software used need to be supplemented in the paper.
- The writing style of some content does not conform to the writing rules of the Agriculture journal.
- In the content, some abbreviations are not fully defined. Such as INEGI, SEGOB, GUI, MDD, 3D, 2D, UAV, AUC, etc.
- Mathematical symbols are not fully defined and explained
- The position of the figuret in the text must be after the text of the figure.
- It is recommended that the numerical display format of the figure should be unified, such as whether to use scientific notation (Figs. 5 and 11). The unit indicator can be moved up and below the axis of the chart, which will look better.
Fig 6 should place what the color change represents, such as height.
Fig. 12: Could you please explain the deviations in three cases?
Page 2, line 81-83: I believe that this part is needless, as you refer to the content of the paper in the previous paragraph. If you think appropriate, please add some part of it in the previous paragraph.
Page 10, line 330: Please explain what is agrolite, the cotidellons?
Page 13, line 386: Please elaborate and discuss Figure 11.
Page 16, line 405: A space is required between GMM and method, "GMMmethod".
Page 16, line 407: Remove Decision Tree Classifier, use DCT only.
Page 3, line 103: The growth appearance shown in Figure 1 does not match the description of the figure, please confirm. Is it “class” or “stage”?
Author Response
This paper explains designing and implementation of a dynamic measurement system using Kinect 2.0 services. The author utilizes the GMM, Mean Shift, DTC to achieve automatic segmenting and classifying portos tomato seedlings. This study also illustrates the crop production accuracy of the growth stage obtained by each classifier compared to the crop production accuracy of experts. Experimental studies like this one, which also involve construction and experiments, require a significant effort and investment of time and resources. Overall the paper is logically structured and the work is presented in a reasonable manner. The conditions for dynamic measurement system evaluation were appropriately designed. However, I would like to see a more in-depth description of the novelty of this work, and how it relates to previous work in the field of phenotyping systems with crop growth mapping and measurement services. The following are some suggestions.
1.- Introductions describes the actuality of the technological solution, but does not reveal the novelty of the question or scientific relevance.
A: Thanks for your comments. We add the novelty explanation in section 1 on lines 23-30 and the discussion section on lines 493-499.
2.- EDO. ARTE. Improve the literature review on segmentation, feature extraction, and classifier applied to dynamic measurement systems, especially the greenhouse environment and greenhouse crops growth. This is required in order to justify the proposed approach. Extend the review to imaging systems applied to model vegetables crop growth.
A: Comment applied. We add section 2, "related works", divided into 2D and 3D systems. First, we include five more cites where the authors used 2D data using Neuronal Networks. The authors acquire the data set in the agronomy environment. Second, the 3D data system section explains 3D segmentation, feature extraction and classification. In this section, we compared our results qualitatively.
Due to the complexity of the data acquisition in agricultural environments, each stage of processing must be adapted to the type of information acquired with the sensor and the seedling's physiology. Finally, please let us know if you think we should add any item(s) that we have not considered.
3.- Please include a nomenclature defining all variables, parameters and abbreviations before the introduction.
A: Comment applied. We added on abbreviations.
4.- Methodology section contain mathematical description of the technological solution (segmentation, feature extraction, and classifying) of production of portos tomato services, but the performance should be quantitatively evaluated so that it can be compared with other works. My point is that any dynamic measurement model of phenotyping system must take into account the existence of those several time-constants. As a matter of fact, one attractive feature of the proposed dynamic measurement architecture is the combination of several sub-model (segmentation, extraction and classification) and authors should connect their design with the time-scales found in the greenhouse phenotyping and measurement system. My suggestion is for authors to use short sampling or scan times (Kinect v2) to check the performance and behavior of their dynamic measurement systems.
A: In this study, we rely on the quality processes established by the greenhouse for tomato monitoring. The objective is to imitate their internal processes, proposing an efficient, effective and low-cost system to make decisions in the use of agrochemicals and water and predict the final production. For this reason, the study does not contemplate time series monitoring. However, your valuable comment will be taken into account for future work and develop a more general system capable of measuring other types of plants where the dynamic model is part of the internal process of the greenhouse.
5.- Please don't start the paragraph with a sentence that just says the Figure shows specific information - you should use the main text to explain what you found in your research, then cite a Fig that provides explanatory details.
A: Comment applied.
6.- Dynamic measurement system with Kinect v2 services itself is presented in section 3, 4, amd 5 in detail, however, it does not solve or analyze any of scientific uncertainties or authors did not disclose that in the manuscript.
A: The study shown in this article proposes a new methodology for monitoring the growth and development of porto tomato seedlings. The system can predict crop production from the 15th day. Trained personnel divide the seedlings into three stages of growth and failure. Based on experience, they show that the 15th day after planting is the ideal day for nutrient application and irrigation decisions. An economical, easily transportable system is proposed without modifying greenhouse conditions. The average greenhouse error for yield estimation is $\pm2\%$. In our study, we reduced this error by up to $1\%$. We evaluate optimising and comparing various methods at each critical stage, segmentation and grading, to determine which is most efficient under the given conditions. The system will monitor $100\%$ of seedlings without fatigue in the future. Finally, uncertainties are not expressed because they are considered intrinsically in the decision parameters.
7.- The way you present Result is a bit strange. You should present all measurements and analysis methods of the experiment first (GMM, DTC, etc.) and then the experimental set-up. Here you miss the analysis of the methodology to measure the surface of the leaves (stem of height) in points, the number of the leaves and so on and how you determined the optimum values presented in Fig. 9.
A: Comment applied. In the experimental set-up, it moved to methodology. Lines 414-421 explains the seedling height. Lines 423-430 illustrates extracting seedling development using MeanShift. In both cases, the idea was rewritten. Finally, we added a paragraph on lines 473-484 to explain optimisation thresholds.
8.- System implementation. Authors ought to provide some details of how the proposed dynamic measurement systems was programmed in a computer (GUI) system.
A: We add the explanation on lines 527-531.
9.- as Discussion is too short, it should be rewritten in accordance to results of statistical analyses (especially in terms of computation load and time with different dynamic measurement system) and en-reached about several recent works. It should be noted that you need to use references to explain your results and discuss them when you are trying to explain something found and not presented in previous work. The content of Section 6.1 is nothing new and does not show some scientific uncertainties and discoveries.
A: Comment applied. The discussion was broadened, adding computational time, measures error, and comparison with similar research in section 6.1 (old version). The authors of cite 96 describe the difficulty in finding a quantitative comparison criterion because the approaches focus on different tasks.
10.- The author does not describe in detail how to obtain the comparative data in Tables 3 and 4, including the indicators that define the performance evaluation (crop production accuray and model accuracy)?
A: We add the explanation for Tables 3 and 4 on lines 442-448 and 449-454, respectively.
11.- The authors do not describe in detail how the Kinect platform works in the greenhouse, including how the hardware device performs large-scale scanning work, how it moves, and what conditions are required?
A: We add the explanation on lines 44-52.
12.- Although the author has described the theory required in the proposed method, the implementation details of the six stages and the names of the processing software or statistical software used need to be supplemented in the paper.
A: We answered in section discussion on lines 531-533 and 543-549.
13.- The writing style of some content does not conform to the writing rules of the Agriculture journal.
A: We change research manuscript sections: 1) Introduction, 2) Related Works, 3) Materials and Methods, 4) Results, 5) Discussion, 6) Conclusions (optional).
14.- In the content, some abbreviations are not fully defined. Such as INEGI, SEGOB, GUI, MDD, 3D, 2D, UAV, AUC, etc.
A: Comment applied, see abbreviations.
15.- Mathematical symbols are not fully defined and explained.
A: Thanks for your observation. We are revised and added the definition of missing symbols on the lines 248, 299, 304, 306, 316, 319-327, 335, 337-339, 366, 369.
16.- The position of the figure in the text must be after the text of the figure.
A: Comment applied. However, with Table 5 and Figures 12 and 13. We need to remove the lines and comments of the reviewers.
17.- It is recommended that the numerical display format of the figure should be unified, such as whether to use scientific notation (Figs. 5 and 11). The unit indicator can be moved up and below the axis of the chart, which will look better.
A: Thank you for your comment. We have taken it into account in lines 403, 377 and Figures 4 and 9.
18.- Fig 6 should place what the color change represents, such as height.
A: Comment applied. See line 378.
19.- Fig. 12: Could you please explain the deviations in three cases?
A: Comment applied. Lines 513-521.
20.- Page 2, line 81-83: I believe that this part is needless, as you refer to the content of the paper in the previous paragraph. If you think appropriate, please add some part of it in the previous paragraph.
A: Comment applied. We removed lines 41-43.
21.- Page 10, line 330: Please explain what is agrolite, the cotidellons?
A: Comment applied. We add both definitions on lines 261-265. We believe it will help the reader in which we have placed both definitions in section 3.
22.- Page 13, line 386: Please elaborate and discuss Figure 11.
A: Thank you for your comment. It was taken into account on lines 422-430.
23.- Page 16, line 405: A space is required between GMM and method, "GMMmethod".
A: Comment applied on the line 525.
24.- Page 16, line 407: Remove Decision Tree Classifier, use DCT only.
A: Comment applied on the line 526.
25.- Page 3, line 103: The growth appearance shown in Figure 1 does not match the description of the figure, please confirm. Is it “class” or “stage”?.
A: Comment applied on the lines 187-194.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
This paper has been revised and is now better structured and more readable. The paper may be accepted subject to a thorough language check. See lines 414 and 560-561 for examples.
Author Response
The orange colour shows comments on reviewer 1, the colour blue shows comments on reviewer 2, and the green colour shows comments on both reviewers.
This paper has been revised and is now better structured and more readable. The paper may be accepted subject to a thorough language check. See lines 414 and 560-561 for examples.
A: Thanks for your comments. We made the appropriate changes to the language.
Author Response File: Author Response.pdf
Reviewer 2 Report
The authors have greatly improved the quality and impact of their paper. A few minor revision are list below.
-Some source in References list are not according to the requirements.
- Please unify the whole work and take the past tense in the description of the results and discussion.
Author Response
The orange colour shows comments on reviewer 1, The colour blue shows comments on reviewer 2, and the green colour shows comments on both reviewers.
The authors have greatly improved the quality and impact of their paper. A few minor revision are list below.
Some source in References list are not according to the requirements.
A: Comment applied. We have changed the references list. Now, according to the requirements, can you see blue colour changes.
Please unify the whole work and take the past tense in the description of the results and discussion.
A: Thanks for your comments. We change to the past tense the results and discussion.
Author Response File: Author Response.pdf
