Using a Bunch Testing Time Augmentations to Detect Rice Plants Based on Aerial Photography

Round 1

Reviewer 1 Report

In this paper, the author proposed a grid cropping method, a multi-scale prediction method, and the other augmentations to enhance the regular detector’s ability to accurately capture tiny crops. Additionally, an efficient object detector was introduced to expedite the detection process. The paper is well thought out and worked on. But I also have some suggestions. 

1) What does the first paragraph of Section 2 mainly explain? The author should give a title of this paragraphy.

2) The author simply provided Figures 4 and 5, without providing any explanation in the paper.

3) In the Section 4.2, the author should provide more comparison results with other state-of-the-art methods to evaluate the performance of the proposed method.

4) The author proposed many methods, i.e. a grid cropping method, a multi-scale prediction method, an efficient object detector and so on, but there is no clear framework for the relationship between these methods.

5) Reviewing throughout the paper, the English grammar, spelling, and sentence structure still needs improve. 

 Minor editing of English language required

Author Response

Thank you for generously providing so much guidance. Our paper will benefit from these thoughtful suggestions.

Q1. What does the first paragraph of Section 2 mainly explain? The author should give a title of this paragraph.
A1. We have added the title: "Machine Learning on Agriculture Application."

Q2. The author simply provided Figures 4 and 5, without providing any explanation in the paper.
A2. We have added appropriate explanations in the relevant sections.

Q3. In Section 4.2, the author should provide more comparison results with other state-of-the-art methods to evaluate the performance of the proposed method.
A3. We have conducted new ablation experiments and enriched the content with more analyses.

Q4. The author proposed many methods, i.e. a grid cropping method, a multi-scale prediction method, an efficient object detector and so on, but there is no clear framework for the relationship between these methods.
A4. We have redrawn Figure 5 to provide a more comprehensive framework and supplemented the corresponding section with a detailed explanation of the proposed TTA.

Q5. Reviewing throughout the paper, the English grammar, spelling, and sentence structure still need improvement.
A5. Thank you for the reminder. We have made every effort to enhance grammar, spelling, and sentence structure.

Author Response File: Author Response.pdf

Reviewer 2 Report

Lines 1 to 9:The abstract did not provide a brief description of the enhancement methods used during the testing process.

Lines 4 to 5: I think this manuscript should reorganize the content of the abstract, especially the pointed-out section, to clearly express the content, rather than using expressions like "other enhancement methods".

Lines 10: Is there any relevant content in the text for "3D point clouds" and "VGG" in the keywords? The purpose of the keywords should be clear, rather than adding content casually.

Lines 24 to 29:this manuscript mentioned that CNN faces significant limitations in the field of agriculture, citing factors such as soil composition, climate conditions, and sunlight exposure, but not related to the issues addressed in this paper.

Lines 45 to 47: this manuscript showed that network pruning methods, multi-scale prediction methods, and other methods were introduced during the testing process. What other methods refer to? I didn't explain clearly.

Lines 184 to 186:In Figure 4, this manuscript mentioned the overall architecture of the enhanced CSL-YOLO, which has made some changes compared to CSL-YOLO. It is best to add a comparison model.

Lines 186: In Figure 4, this manuscript should clearly express what the different modules are in the network structure.

Lines 219: The content in Figure 5 should be further supplemented, and the output of the red arrow should be indicated.

Lines 220 to 221:this manuscript showed setting R to 3 is a balance between speed and accuracy, without relevant charts or data support.

Lines 232 to 236: The abbreviation "Testing Time Augmentation (TTA)" should have a suitable position in the article, and various abbreviations such as "TTA" in Figure 5 should be made clear to people.

Lines 245 to 255: The grid in Figure 6 cannot express this work well. Is there a better way to showcase the work.

Lines 316: The ablation experiment section lacks workload, please try to supplement it as much as possible.

Lines 360 to 367:this manuscript mentioned that the combination of various time enhancement methods, including TTCG MSF MSF, is not completely consistent with the one mentioned earlier, so maybe miss one point.

In this winter. I was glad to read this manuscript:“Using a Bunch Testing Time Augmentations to Detect Rice Plant Based on Aerial Photography”.In this paper, due to the small size of the target crops and the high image resolution, it is difficult for conventional detection programs to accurately identify them. However, there were some inconsistencies between the previous and subsequent sections in this paper. Reconsider after major revision

Minor editing of English language required

Author Response

Thank you for generously providing so much guidance. Our paper will benefit from these thoughtful suggestions.

Q1. Lines 1 to 9: The abstract did not provide a brief description of the enhancement methods used during the testing process.
A1. We have revised the abstract based on the suggestions.

Q2. Lines 4 to 5: I think this manuscript should reorganize the content of the abstract, especially the pointed-out section, to clearly express the content, rather than using expressions like "other enhancement methods".
A2. We have restructured the abstract as recommended.

Q3. Lines 10: Is there any relevant content in the text for "3D point clouds" and "VGG" in the keywords? The purpose of the keywords should be clear, rather than adding content casually.
A3. We have rewritten the keywords based on the suggestions.

Q4. Lines 24 to 29: This manuscript mentioned that CNN faces significant limitations in the field of agriculture, citing factors such as soil composition, climate conditions, and sunlight exposure, but not related to the issues addressed in this paper.
A4. We have removed unrelated issues, focusing on the characteristics of the AI Cup 2021 dataset, and added two demo images with significant differences in lighting conditions in F1 Score.

Q5. Lines 45 to 47: This manuscript showed that network pruning methods, multi-scale prediction methods, and other methods were introduced during the testing process. What other methods refer to? I didn't explain clearly.
A5. We have rephrased the abstract and introduction according to the suggestions and provided more details about the proposed TTA where necessary.

Q6. Lines 184 to 186: In Figure 4, this manuscript mentioned the overall architecture of the enhanced CSL-YOLO, which has made some changes compared to CSL-YOLO. It is best to add a comparison model.
A6. We have redrawn Figure 4, emphasizing the modified parts.

Q7. Lines 186: In Figure 4, this manuscript should clearly express what the different modules are in the network structure.
A7. We have added details to Figure 4 and included corresponding details in the text.

Q8. Lines 219: The content in Figure 5 should be further supplemented, and the output of the red arrow should be indicated.
A8. We have redrawn Figure 5 according to the suggestions.

Q9. Lines 220 to 221: This manuscript showed setting R to 3 is a balance between speed and accuracy, without relevant charts or data support.
A9. We have emphasized the definition of R in Figure 5 and the corresponding text.

Q10. Lines 232 to 236: The abbreviation "Testing Time Augmentation (TTA)" should have a suitable position in the article, and various abbreviations such as "TTA" in Figure 5 should be made clear to people.
A10. Figure 5 has been redrawn based on the suggestions.

Q11. Lines 245 to 255: The grid in Figure 6 cannot express this work well. Is there a better way to showcase the work.
A11. Figure 6 has been redrawn.

Q12. Lines 316: The ablation experiment section lacks workload; please try to supplement it as much as possible.
A12. We have added new ablation experiments and enriched the content for a more detailed analysis.

Q13. Lines 360 to 367: This manuscript mentioned that the combination of various time enhancement methods, including TTCG MSF MSF, is not completely consistent with the one mentioned earlier, so maybe miss one point.
A13. We have rewritten the corresponding text based on the suggestions and added an abbreviation table at the end.

Author Response File: Author Response.pdf

Reviewer 3 Report

Authors proposed a low resolution rice plant detection algorithm using a multi-scale grid cropping method. Following issues need addressing:
- The computation time for the developed system has not been reported. The authors should include this information.
- Please explain the term 'multi-sclae prediction', does it mean the multi-resolution method? Further elaboration is required on this in paper. It is creating quite confusion in paper.
- The methodology section need to be improved, in this regard, an algorithm or pseudocode can be written.
- More qualitative detection results may be visualized.
- The actual novelty in the work is not visible, please discuss this in the contribution.
- In literature, discuss these works: Singh et al. (2023) Latent Graph Attention for Enhanced Spatial Context. Biswas et al. (2023). pNNCLR: Stochastic pseudo neighborhoods for contrastive learning-based unsupervised representation learning problems.
- The model diagram is not explained well. Please add the corresponding images/illustrations to the text. Also explain what the arrows mean, what is input output?
- Results section may be expanded using some more metrics like mIoU or PSNR, more ablation studies, and other datasets, with detailed analysis.

None

Author Response

Thank you for generously providing so much guidance. Our paper will benefit from these thoughtful suggestions.

Q1. The computation time for the developed system has not been reported. The authors should include this information.
A1. We have incorporated new ablation experiments, investigating the variations in F1 and FLOPs.

Q2. Please explain the term 'multi-scale prediction,' does it mean the multi-resolution method? Further elaboration is required on this in the paper. It is creating quite confusion in the paper.
A2. We have redrawn Figure 5 to provide details on MSP and other TTA, and have also added more content to explain the operational principles of MSP.

Q3. The methodology section needs to be improved; in this regard, an algorithm or pseudocode can be written.
A3. We have done our best to enhance the methodology section.

Q4. More qualitative detection results may be visualized.
A4. We have added six detection result images in different environments to the conclusion chapter.

Q5. The actual novelty in the work is not visible; please discuss this in the contribution.
A5. We have emphasized the contribution in the introduction and conclusion as per the suggestions.

Q6. In the literature, discuss these works: Singh et al. (2023) Latent Graph Attention for Enhanced Spatial Context. Biswas et al. (2023). pNNCLR: Stochastic pseudo neighborhoods for contrastive learning-based unsupervised representation learning problems.
A6. We have added more discussion and citations as recommended, referencing these two papers appropriately.

Q7. The model diagram is not explained well. Please add the corresponding images/illustrations to the text. Also, explain what the arrows mean, what is input-output?
A7. We have redrawn Figure 4 and rewritten captions and corresponding content to provide more details on Enhanced CSL-YOLO. Additionally, Figure 5 has been redrawn to offer a more comprehensive framework.

Q8. The Results section may be expanded using some more metrics like mIoU or PSNR, more ablation studies, and other datasets, with detailed analysis.
Due to the lack of corresponding labels for the AI CuP 2021 dataset, which only provides root coordinates for each rice plant in aerial images, it is challenging to calculate other metrics. Additionally, we have expanded the ablation experiments in the experimental section and included more analysis in the text.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Author Response

Thank you for your generous guidance. Our paper will greatly benefit from these thoughtful suggestions.

Author Response File: Author Response.pdf

Reviewer 2 Report

Lines 156: I think for the professional term "FLOPs", the full name should also be used before using abbreviations.

Lines 186: In Figure 4, what you want to show is your network framework, and you need to add the names of different components, not highlight the parts you added.

Lines 380-396: In this section, you compared the enhanced YOLO, but the effect was not significant. The fusion of enhanced YOLO and the method you proposed improved the results. You should describe the reasons as comprehensively as possible in the article.

Minor editing of English language required

Author Response

Thank you for your generous guidance. Our paper will greatly benefit from these thoughtful suggestions.

Q1. Lines 156: I think for the professional term "FLOPs," the full name should also be used before using abbreviations.
A1. We have revised the term as suggested.

Q2. Lines 186: In Figure 4, what you want to show is your network framework, and you need to add the names of different components, not highlight the parts you added.
A2. CSL-YOLO consists of CSL-Bone (backbone), CSL-FPN (neck), and YOLO head (detection head). We have modified the backbone and neck components. The caption for Figure 4 has been rewritten according to your suggestion, providing a clearer representation of the network framework.

Q3. Lines 380-396: In this section, you compared the enhanced YOLO, but the effect was not significant. The fusion of enhanced YOLO and the method you proposed improved the results. You should describe the reasons as comprehensively as possible in the article.
A3. We have restructured this section based on your suggestions, providing a more detailed description of the challenges addressed by each method in the proposed TTA.

Author Response File: Author Response.pdf

Reviewer 3 Report

Most of the suggestions have been incorporated by the authors, the paper has improved its quality and may be accepted now.

Author Response

Thank you for your generous guidance. Our paper will greatly benefit from these thoughtful suggestions.

Author Response File: Author Response.pdf