Algorithm for Corn Crop Row Recognition during Different Growth Stages Based on ST-YOLOv8s Network

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper proposes an approach for recognizing crops under different growth stages. Here are some considerations.

1. The authors should discuss properly the advantages and disadvantages shown in Table 1 for the proposed navigation methods. In other words, the given description is too broad and should be further detailed.

2. The paper lacks relevant state-of-the-art on newer applications of YOLO in the agricultural field. For example, check https://doi.org/10.1007/s11042-023-16570-9 or https://doi.org/10.1016/j.compag.2024.108728.

3. The dataset description can be improved by including the overall number of labels associated with crops. Furthermore, the authors should consider publishing their dataset for reproducibility. Finally, the description lacks information on the sensor used (only the "RGB camera" is insufficient).

4. The authors should also provide a comparison in terms of mAP 0.5-0.95. Furthermore, it would be interesting to understand the computational cost in terms of FPS imposed by ST-YOLOv8.

5. The authors should provide further details on the interpretation provided by the segmentation method. Furthermore, what happens when lighting variations are within the scene? Finally, why did the authors not consider the use of deep learning models for segmentation, such as SAM or, even better, ontology-based ones, such as GroundedSAM? It would be interesting to provide a full comparison, even in qualitative terms.

6. Furthermore, as the authors provided only a subset of optimal images within the comparison, a set of more challenging ones should be proposed and discussed, i.e., with occlusions, non-optimal crop alignment, etc., to validate the methodology.

For all these reasons, I suggest the paper goes through a major revision before being considered for publication.

Author Response

Dear reviewer, I am deeply grateful for your detailed review of my manuscript and for providing valuable suggestions. Your advice is crucial for improving the quality and rigor of my paper, and I have been greatly inspired by it. I have carefully studied each of your suggestions and made comprehensive and meticulous revisions to the manuscript. The revised content has been marked in red for your quick reference to each change. In the process of revision, I have tried my best to ensure that every one of your suggestions has been fully considered and implemented, in order to make the structure of the paper more reasonable, the argumentation more rigorous, and the expression more clear.

Comments 1: [The authors should discuss properly the advantages and disadvantages shown in Table 1 for the proposed navigation methods. In other words, the given description is too broad and should be further detailed.]

Response 1: [Thank you so much for carefully reviewing my manuscript and offering valuable insights. Regarding the discussion of the advantages and disadvantages of the proposed navigation methods in Table 1, we realized that the original description was overly general and lacked specific analysis. Therefore, we made significant revisions and additions between lines 53 and 70, providing a more detailed description and discussion of each navigation method. Our goal is to offer more specific explanations and comparisons to help readers gain a more comprehensive understanding of the strengths and weaknesses of the various navigation methods.]

Comments 2: [The section "Dataset construction" mentions two growth stages, howerver, I think those are imprecise. You should specify the stages according vegetative stages of corn crop (i.e. V0 , ... V8, .. Vn)]

Response 2: [Thank you so much for your valuable suggestions. In constructing the dataset, we indeed focused primarily on two critical corn growth stages because our collection efforts were centered on supporting the application of corn spraying robots. Your feedback has highlighted to us that such a division may not fully capture the continuous state of corn growth, which is of significant guidance for our future research. We are extremely grateful for your insight into our future direction. In our upcoming work, we will continue to monitor all stages of corn growth and strive to collect more diverse and precise data to reflect the entire corn growth process more comprehensively.]

Comments 3: [The photographs  seem perfectly framed, did you consider special cases such as a single row of crops or different number of corn plants in the row?]

Response 3: [Thank you very much for your understanding and feedback on our work. We have indeed noticed the special cases you mentioned, such as single-row crops or crop rows with varying numbers of corn plants. Since we mainly collect data in standardized farmland, these special cases are relatively rare. However, we are fully aware of their significant impact on the model's performance in practical applications. Therefore, we are continuously working to collect a more diverse dataset, including special cases like occlusion, suboptimal alignment, single-row crops, curved crop rows, and broken rows. Our next research will delve deeper into these special cases and explore appropriate solutions. We have updated the relevant content in the manuscript, and you can find it in lines 417 to 420.]

Comments 4: [For the ST-YOLOv8s Neural Network (ANN), What about training curves or training performance? How does anyone know if the ANN were trained properly? What about training curves or training performance? How does anyone know if the ANN were trained properly?]

Response 4: [Thank you so much for your valuable opinions and suggestions during the review of our paper. To further enhance the persuasiveness and readability of the paper, we have redesigned the content of Section 3.2, including adjustments to Figures 7 and 8, to provide a more intuitive presentation of our results. Additionally, we have added Figure 9 and relevant descriptions of the training curves to give readers a deeper understanding of the training process and performance of our algorithm. These modifications aim to provide richer and more intuitive information, helping readers to fully understand the content and methodology of our research. We firmly believe that these improvements will further enhance the quality and academic value of the paper.]

Comments 5: [In The "Improved supergreen method" section the changes should be shown with histrograms.]

Response 5: [Thank you so much for your valuable opinions and suggestions during the review of our paper. In the research on crop row localization and fitting, we employed a parameterized hypergreen method, combined with a local-to-global detection approach and least squares method, to achieve more precise crop row identification. Since the overall performance of the algorithm is primarily reflected in the final crop row detection results, it is difficult to directly assess the effectiveness of the improved hypergreen method using data alone. However, we have detailedly showcased the image results processed by the hypergreen method in the manuscript and visually presented the effects of parameter adjustments through visualization methods. These visual images not only help us deeply understand the performance of the hypergreen method, but also provide readers with an intuitive understanding, enabling them to more clearly comprehend our research methods and achievements.]

Comments 6: [Crop row segment detection experiment" section should contain an explanation of the metrics, the number of images used, values of P and R.]

Response 6: [Thank you very much for your valuable opinions and suggestions during the review of our paper. We have carefully considered and incorporated your feedback, making corresponding revisions and improvements to the paper. In the section on crop row segmentation and detection, we have thoroughly supplemented detailed explanations of the evaluation metrics, which are crucial for comprehensively assessing the performance of crop row detection algorithms. We have explicitly explained the significance of MAP, F1 score, precision (P), and recall (R), as well as the role of category serial number i and total number of categories n in the evaluation process. Clarifying the meanings of these parameters will help enhance the readability and persuasiveness of the paper, enabling readers to more fully understand our research results and evaluation methods. We have updated the relevant content in the manuscript, which you can find in lines 291 to 296.]

Comments 7: [It could be interesting to know about the type or characteristics of robot your propose to use.]

Response 7: [Thank you very much for your in-depth understanding and valuable feedback on our work. We are deeply honored as your feedback serves as an important guide for our progress. In the future, developing agricultural robots and plant protection drones equipped with efficient visual navigation systems and precise pesticide spraying control systems is a key research direction for our research group. We have updated the relevant content in the manuscript, which you can find in lines 420 to 423.]

Comments 8: [Comments on the Quality of English Language: Capitalize the initial letters of titles or known methods.  Avoid expression like "this paper constructs..", "this paper proposes..", etc. Nomination of Figures shold be the same in caption and reference in text. Use S.I. format  for physical units]

Response 8: [Thank you so much for your thorough review and valuable suggestions on this paper. I deeply agree with them and have made the necessary adjustments accordingly. I believe these modifications will make the article more compliant with academic norms and better convey our research findings. Additionally, I will be more mindful of my choice of expressions in future writing to avoid similar issues. Once again, I appreciate your thoughtful guidance, as your suggestions are crucial to improving the quality of our research.]

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The topic of your manuscript is intereresting, However I consider there have some omissions and mistakes that should be reviewed:

- The Table 1 and paragraphs that refer to this table should contain reference(s).

- The section "Dataset construction" mentions two growth stages, howerver, I think those are imprecise. You should specify the stages according vegetative stages of corn crop (i.e. V0 , ... V8, .. Vn)

- The photographs  seem perfectly framed, did you consider special cases such as a single row of crops or different number of corn plants in the row?

-For the ST-YOLOv8s Neural Network (ANN), What about training curves or training performance? How does anyone know if the ANN were trained properly? What about training curves or training performance? How does anyone know if the ANN were trained properly?

- In The "Improved supergreen method" section the changes should be shown with histrograms.

- "Crop row segment detection experiment" section should contain an explanation of the metrics, the number of images used, values of P and R.

- It could be interesting to know about the type or characteristics of robot your propose to use.

Best regards

Comments on the Quality of English Language

- Capitalize the initial letters of titles or known methods.

- Avoid expression like "this paper constructs..", "this paper proposes..", etc.

- Nomination of Figures shold be the same in caption and reference in text.

-Use S.I. format  for physical units

Author Response

Dear Reviewer, I am profoundly grateful for your thorough review and invaluable feedback. Your suggestions are of utmost importance in enhancing the quality of my paper and have guided me in the right direction for improvement. Pursuant to your guidance, I have made comprehensive and detailed revisions to the manuscript. The changes have been clearly highlighted in red for your convenience to visually track each alteration. During the revision process, I have carefully considered every one of your suggestions and strived to incorporate them into every part of the paper. I am confident that with these revisions, the structure of the paper will be more logical, the arguments will be tighter, and the expression will be clearer. At the same time, I deeply understand that academic research is a continuous process of exploration and progress. Your advice will serve as an important reference and inspiration for my future research work. Once again, I express my heartfelt gratitude for your thoughtful guidance and selfless assistance!

Comments 1: [The Table 1 and paragraphs that refer to this table should contain reference(s).]

Response 1: [Thank you so much for your profound understanding of our research work and for providing valuable feedback. We highly value your suggestions and have carefully revised the paragraphs related to Table 1. We have also added relevant references to ensure that our viewpoints and analysis are based on the latest and most accurate research materials. Once again, we appreciate your valuable advice.]

Comments 2: [The section "Dataset construction" mentions two growth stages, howerver, I think those are imprecise. You should specify the stages according vegetative stages of corn crop (i.e. V0 , ... V8, .. Vn)]

Response 2: [Thank you so much for your valuable suggestions. In constructing the dataset, we indeed focused primarily on two critical corn growth stages because our collection efforts were centered on supporting the application of corn spraying robots. Your feedback has highlighted to us that such a division may not fully capture the continuous state of corn growth, which is of significant guidance for our future research. We are extremely grateful for your insight into our future direction. In our upcoming work, we will continue to monitor all stages of corn growth and strive to collect more diverse and precise data to reflect the entire corn growth process more comprehensively.]

Comments 3: [The photographs  seem perfectly framed, did you consider special cases such as a single row of crops or different number of corn plants in the row?]

Response 3: [Thank you very much for your understanding and feedback on our work. We have indeed noticed the special cases you mentioned, such as single-row crops or crop rows with varying numbers of corn plants. Since we mainly collect data in standardized farmland, these special cases are relatively rare. However, we are fully aware of their significant impact on the model's performance in practical applications. Therefore, we are continuously working to collect a more diverse dataset, including special cases like occlusion, suboptimal alignment, single-row crops, curved crop rows, and broken rows. Our next research will delve deeper into these special cases and explore appropriate solutions. We have updated the relevant content in the manuscript, and you can find it in lines 417 to 420.]

Comments 4: [For the ST-YOLOv8s Neural Network (ANN), What about training curves or training performance? How does anyone know if the ANN were trained properly? What about training curves or training performance? How does anyone know if the ANN were trained properly?]

Response 4: [Thank you so much for your valuable opinions and suggestions during the review of our paper. To further enhance the persuasiveness and readability of the paper, we have redesigned the content of Section 3.2, including adjustments to Figures 7 and 8, to provide a more intuitive presentation of our results. Additionally, we have added Figure 9 and relevant descriptions of the training curves to give readers a deeper understanding of the training process and performance of our algorithm. These modifications aim to provide richer and more intuitive information, helping readers to fully understand the content and methodology of our research. We firmly believe that these improvements will further enhance the quality and academic value of the paper.]

Comments 5: [In The "Improved supergreen method" section the changes should be shown with histrograms.]

Response 5: [Thank you so much for your valuable opinions and suggestions during the review of our paper. In the research on crop row localization and fitting, we employed a parameterized hypergreen method, combined with a local-to-global detection approach and least squares method, to achieve more precise crop row identification. Since the overall performance of the algorithm is primarily reflected in the final crop row detection results, it is difficult to directly assess the effectiveness of the improved hypergreen method using data alone. However, we have detailedly showcased the image results processed by the hypergreen method in the manuscript and visually presented the effects of parameter adjustments through visualization methods. These visual images not only help us deeply understand the performance of the hypergreen method, but also provide readers with an intuitive understanding, enabling them to more clearly comprehend our research methods and achievements.]

Comments 6: [Crop row segment detection experiment" section should contain an explanation of the metrics, the number of images used, values of P and R.]

Response 6: [Thank you very much for your valuable opinions and suggestions during the review of our paper. We have carefully considered and incorporated your feedback, making corresponding revisions and improvements to the paper. In the section on crop row segmentation and detection, we have thoroughly supplemented detailed explanations of the evaluation metrics, which are crucial for comprehensively assessing the performance of crop row detection algorithms. We have explicitly explained the significance of MAP, F1 score, precision (P), and recall (R), as well as the role of category serial number i and total number of categories n in the evaluation process. Clarifying the meanings of these parameters will help enhance the readability and persuasiveness of the paper, enabling readers to more fully understand our research results and evaluation methods. We have updated the relevant content in the manuscript, which you can find in lines 291 to 296.]

Comments 7: [It could be interesting to know about the type or characteristics of robot your propose to use.]

Response 7: [Thank you very much for your in-depth understanding and valuable feedback on our work. We are deeply honored as your feedback serves as an important guide for our progress. In the future, developing agricultural robots and plant protection drones equipped with efficient visual navigation systems and precise pesticide spraying control systems is a key research direction for our research group. We have updated the relevant content in the manuscript, which you can find in lines 420 to 423.]

Comments 8: [Comments on the Quality of English Language: Capitalize the initial letters of titles or known methods.  Avoid expression like "this paper constructs..", "this paper proposes..", etc. Nomination of Figures shold be the same in caption and reference in text. Use S.I. format  for physical units]

Response 8: [Thank you so much for your thorough review and valuable suggestions on this paper. I deeply agree with them and have made the necessary adjustments accordingly. I believe these modifications will make the article more compliant with academic norms and better convey our research findings. Additionally, I will be more mindful of my choice of expressions in future writing to avoid similar issues. Once again, I appreciate your thoughtful guidance, as your suggestions are crucial to improving the quality of our research.]

 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors fixed the highlighted issue, improving the quality of the paper, which can now be considered for publication.

Author Response

Dear Reviewer,

 

Thank you very much for your valuable comments and recognition! We are delighted to learn that after the revisions, you believe we have addressed the issues you highlighted, and the quality of the paper has been improved. This is not only a testament to the hard work we have put in but also an encouragement for us to continue improving.

 

Once again, we sincerely appreciate your thorough review and invaluable suggestions.

 

Best regards

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Dear author,

I received your answers to the questions asked. I think there are some unresolved questions or incomplete answers.

With reference to:

“The section "Dataset construction" mentions two growth stages, howerver, I think those are imprecise. You should specify the stages according vegetative stages of corn crop (i.e. V0 , ... V8, .. Vn)”

- I understand that you did hard work in the field, however it is common for researchers to forget to take some data into account. For example, I consider that the identification of the vegetative state [1] is very important so that the methodology can be reproduced by other agricultural researchers or technical personnel. I think you should review your data set and establish a statistical range on the vegetative stage for the two groups you mentioned

[1 Corn Growth Stages | Integrated Crop Management (iastate.edu)]

With reference to:

“The photographs  seem perfectly framed, did you consider special cases such as a single row of crops or different number of corn plants in the row?”

I think that the diversity of data is necessary to properly train an ANN, with special cases as you mentioned or with different soil or different lighting, otherwise, when the algorithm is embedded in an electronic system, the precision drops considerably. By now it can be enough mentioned the feature of the images. (Note I did not find line 417)

With reference to:

“For the ST-YOLOv8s Neural Network (ANN), What about training curves or training performance? How does anyone know if the ANN were trained properly? What about training curves or training performance? How does anyone know if the ANN were trained properly?”

You mentioned three ANN in the Table 5, what about with the training curves for the remain two ANN? Rapid learning could be due to monotonous information into images.

With reference to:

“In The "Improved supergreen method" section the changes should be shown with histrograms”

What is the hypergreen method? I guess it's the super green method. Well, you wrote "paper improves the traditional supergreen method, which is improved as shown in equation (1)." Then the coefficient 2 changes to 1.9 and is not explained or compared with the traditional method to show the improvement characteristics. It may be seen as an empirical search, but it is always possible to directly or indirectly demonstrate an improvement. Furthermore, equation (1) has not been expressed correctly with the functional format.

With reference to:

“Crop row segment detection experiment" section should contain an explanation of the metrics, the number of images used, values of P and R.”

I think it is not necessary to include any paragraph, if you used a Python package you should mention that P and R  are intermediate calculations of that package. You should review the grammatic of paragraph included.

Best regards

Comments on the Quality of English Language

Dear Authors,

With reference to:

“Comments on the Quality of English Language: Capitalize the initial letters of titles or known methods.  Avoid expression like "this paper constructs..", "this paper proposes..", etc. Nomination of Figures shold be the same in caption and reference in text. Use S.I. format  for physical units”

The authors did not solve the next:

Capitalize the initial letters of titles

Use the International System of Units (SI). I.e. in paragraphs, tables and figures must be 65 ms and not 65ms or must be 2.50 GHz and not 2.50GHZ

You included other errors, in the references the Chicago format or another established by the Journal should be used, a mixture of IEEE and Chicago formats should not be used.

Author Response

Dear Reviewer, 

I would like to express my sincere gratitude for your invaluable suggestions and your patience in pointing out every detail that needed attention. We have followed your instructions meticulously and conducted a comprehensive and meticulous revision of the manuscript. Each issue you highlighted has been taken seriously, and we have corrected them one by one. To facilitate your review and confirmation, we have highlighted the revised sections in red. We believe that through this effort, the quality of the paper has been significantly enhanced, making it more compliant with academic norms and requirements. Once again, thank you for your attention and support to our work. Your hard work and dedication are the driving force behind our continuous progress.

Comments 1: [“The section "Dataset construction" mentions two growth stages, howerver, I think those are imprecise. You should specify the stages according vegetative stages of corn crop (i.e. V0 , ... V8, .. Vn)”]

Response 1: [Thank you for your suggestions. Following your guidance, we have established a statistical range for the nutritional stages based on our dataset. The images collected in our dataset cover the seedling stage from V4 (when the fourth true leaf is fully expanded) to V6 (when the sixth true leaf is fully expanded), as well as the mid-growth stage from V8 (when the eighth true leaf is fully expanded) to V12 (when the twelfth true leaf is fully expanded). We have included relevant content in Section 2.1, and this addition will provide a reference basis for other agricultural researchers and technicians.]

Comments 2: [“The photographs seem perfectly framed, did you consider special cases such as a single row of crops or different number of corn plants in the row?”]

Response 2: [Thank you very much for your attention to our research. We fully appreciate your concern about the potential decline in algorithm performance under special circumstances, such as varying numbers of corn plants arranged in different patterns. As you rightly pointed out, ensuring the diversity and representativeness of data is crucial when training neural networks. In our current algorithm design and experimental stages, we primarily focused on evaluating performance under standard conditions and did not fully consider various special scenarios. We acknowledge the limitations of our current research and plan to enrich the relevant dataset as much as possible in future work, while optimizing the algorithm specifically for those special cases. We have added this point to the last paragraph of the conclusion section, which you can find in the manuscript. Thank you again for your insightful feedback.]

Comments 3: [“For the ST-YOLOv8s Neural Network (ANN), What about training curves or training performance? How does anyone know if the ANN were trained properly? What about training curves or training performance? How does anyone know if the ANN were trained properly?”]

Response 3: [Thank you very much for your valuable suggestions. We have comprehensively optimized and updated Figure 9 accordingly. In this revision, we have not only included the training results of the ST-YOLOv8s, ST-YOLOv5s, and ST-YOLOv7 networks but also meticulously adjusted the relevant information in the chart to ensure that all details are presented clearly and intuitively. We believe that this improvement will significantly enhance the visualization of the experimental process, enabling viewers to more directly understand the performance of different models during training and their differences.]

Comments 4: [“In The "Improved supergreen method" section the changes should be shown with histrograms”]

Response 4: [Thank you very much for your valuable suggestions, which we have actively taken and put into practice. In Figure 5, we have presented a comparison of the processing effect diagrams between the improved supergreen method and the original supergreen method, and adjusted the expression of related content. Through intuitive image display, readers can instantly capture the significant changes brought about by the improved solution.]

Comments 5: [“Crop row segment detection experiment" section should contain an explanation of the metrics, the number of images used, values of P and R.”]

Response 5: [We are extremely grateful for the detailed and insightful suggestions you have provided, which have significantly contributed to enhancing the quality of our paper. In response to your feedback, we have revised the evaluation metrics in Section 3.2. Specifically, we have added Formulas 7 and 8, which are crucial for elucidating the calculation methods and logic behind our evaluation metrics. We believe that with these modifications, readers will be able to gain a deeper understanding of how we have quantified and evaluated our research results, thereby strengthening the credibility of our conclusions.]

Comments 6: [“Comments on the Quality of English Language: Capitalize the initial letters of titles or known methods.  Avoid expression like "this paper constructs..", "this paper proposes..", etc. Nomination of Figures shold be the same in caption and reference in text. Use S.I. format  for physical units”, "In the references the Chicago format or another established by the Journal should be used, a mixture of IEEE and Chicago formats should not be used"]

Response 6: [Thank you very much for your meticulous attention to the specific errors in our paper. We have conducted a comprehensive review and correction of the title, units, and reference formatting. We fully understand that these details are crucial to the professionalism and rigor of our paper, and we have taken each of your suggestions very seriously. Through this revision, we hope to present a more standardized and accurate paper that better meets academic requirements and provides a smoother reading experience for our readers.]

Best regards

Author Response File: Author Response.pdf