YOLOv7-MA: Improved YOLOv7-Based Wheat Head Detection and Counting
Round 1
Reviewer 1 Report
This research paper is a comprehensive and informative piece of work that advances the field of wheat head detection and counting. It successfully presents the advantages and performance of the YOLOv7-MA model. While the research is valuable and well-executed, certain aspects could be enhanced for better clarity, organization, and understanding.
Abstract:
1. Abbreviations and technical jargon like [email protected], FPS, and R2 should be clarified at their first usage to cater to a broader audience and enhance comprehension.
2. The abstract, while detailed, is rather verbose. Aim for conciseness without omitting crucial information, as an abstract should be a brief summary of the research work.
3. The comparison between the YOLOv7-MA model and other models like Faster-RCNN, YOLOv5, YOLOX, and YOLOv7 should be more succinct and to the point.
4. The statement regarding the superior performance of the model under conditions of low illumination, blur, and occlusion could be strengthened with specific quantitative results.
5. Lastly, the abstract could mention potential applications of this work and propose directions for future research.
Introduction:
1. The addition of a summarizing paragraph outlining the structure of the rest of the paper would enhance the flow and organization of the introduction.
2. The section on contributions would be clearer and more easily understood if the contributions were listed as bullet points.
3. It is recommended to include subsections in the introduction to clearly outline different aspects, such as motivation, contribution, and organization.
Results:
1. A discussion regarding the trade-offs encountered while implementing the new method would enrich the results section. It would give a more comprehensive understanding of the model's capabilities, limitations, and implications.
Overall, the research presents an essential contribution to the field of wheat head detection. The above recommendations are aimed at refining the presentation and organization of the information to ensure the maximum impact of the research findings.
Author Response
尊敬的审稿人，
我们非常感谢您对本稿件的仔细审阅以及您提供的宝贵专业反馈。我们仔细考虑了这些建议，并对我们的稿件进行了逐点修改。同时，我们检查了这份手稿的图形和语法。非常感谢。
此致敬意
Author Response File: 
Author Response.pdf
Reviewer 2 Report
In this study, a new version of the YOLOv7 algorithm developed for wheat head detection and counting is proposed. Deep learning-based plant detection is a current and research-open field. The manuscript is generally well-prepared and explained. Current studies are included in the literature. In addition, the mathematical background of the methods is adequately presented. Here are some suggestions for improving the work:
1) The need in the literature should be explained in a way that emphasizes the original contribution of the study. In particular, existing studies should be presented critically.
2) Why didn't you apply the Faster-RCNN, YOLOv5, YOLOX, and YOLOv7 models in your own dataset? Could have given more ideas for comparison.
3) Although the evaluation metrics in Table 2 and Table 3 were developed with the proposed method, is the amount of increase statistically significant? In this regard, it is recommended to investigate whether the results are significant with a statistical method (McNemar etc.).
4) In the Conclusion section, suggestions for future studies should be added.
Minor editing of English language required
Author Response
Dear Reviewer，
We greatly appreciate your careful review of this manuscript and the valuable professional feedback you have provided. We have carefully considered these suggestions and make changes in our manuscript by point to point. Meanwhile, we have checked the Figures and grammar of this manuscript. Thanks very much.
Best regards,
Author Response File: Author Response.pdf
Reviewer 3 Report
- The authors need to establish a better motivation to their work.
- The citation number should be seprated from the text body (e.g., population[1] --> population [1])
- For the purposes of this study, figure 2 is not really needed for this type of article and journal.
- Similarly, the labeling process is a standard process and the authors should focus on the contribution. Figure 4 is not needed.
- What is the point of table 1, there is no point in mentioning the file names, the authors should have mentioned the number of images in each subset.
- Did the augmentation increase the number of images? This is not clear in the text. This statement is troubling "To obtain accurate detection results with deep learning technology, it is usually necessary to use a training dataset with rich scenes, large scale, and accurate labeling. This study uses data augmentation strategies to process the collected high-resolution wheat images to address the issue of insufficient samples that may cause network overfitting
and enable the model to have stronger generalization ability." Increasing the size of the dataset via augmentation will result in data leaking and inflated good results.
- How was the counting performed from object detection?
- Provide the precsion-recall curves.
- The evaluation method is not clear, did you use cross-validation?
- Similar works in using AI object detection methods (e.g., Faster-RCNN) can be discussed, see Khasawneh, N., Fraiwan, M. & Fraiwan, L. Detection of K-complexes in EEG waveform images using faster R-CNN and deep transfer learning. BMC Med Inform Decis Mak 22, 297 (2022). https://doi.org/10.1186/s12911-022-02042-x
- The table of abbreviations is missing but maybe required by the journal.
Author Response
Dear Reviewer，
We greatly appreciate your careful review of this manuscript and the valuable professional feedback you have provided. We have carefully considered these suggestions and make changes in our manuscript by point to point. Meanwhile, we have checked the Figures and grammar of this manuscript. Thanks very much.
Best regards,
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
The authors addressed my comments.
