SAW-YOLO: A Multi-Scale YOLO for Small Target Citrus Pests Detection
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe paper presents a modification of Yolov8 to better handle the identification of pest of 13 different classes and having three different size categories. Overall the presented idea is good, well experimentation and I have a few comments to improve it:
1. You need to add the contributions of this manuscript in the ending on section 1 before introducing what is in the subsequent sections.
2. The description of the data collection lacks an identification of which class size does each species belong too. And a total count of small, medium, and large to reflect the results in the last analysis.
3. Quality of figure 6 needs to be improved
4. Equation 9 has a typo
5. The caption of figure 7 needs to have more details about the different charts.
6. How well was the identification per species. You showed per size yet not per species.
Author Response
Comments 1: You need to add the contributions of this manuscript in the ending on section 1 before introducing what is in the subsequent sections.
Response 1: Thank you for pointing this out! I agree with this comment. Therefore, I summarize the manuscript's contribution in lines 68-73, indicating how SAW-YOLO is used and summarizing the effect.
Comments 2: The description of the data collection lacks an identification of which class size does each species belong too. And a total count of small, medium, and large to reflect the results in the last analysis.
Response 2: Agree. I have, accordingly, revised the articulation of the division of the three scales of pests in the data collection and added the small, medium, and large scale population divisions to the table to emphasize this point, which can be found in lines 161-164 of the article and table 1.
Comments 3: Quality of figure 6 needs to be improved
Response 3: Agreed, image 6 is low in pixels, we've updated the image in the article to make it look clearer, specifically on page 9 of the article.
Comments 4: Equation 9 has a typo
Response 4: I apologize for such a careless mistake and have corrected the precision "recision" to "precision" on page 11.
Comments 5: The caption of figure 7 needs to have more details about the different charts.
Response 5: Agreed, we didn't express ourselves clearly enough. The description of image 7 has been added to lines 439-446 of the article.
Comments 6: How well was the identification per species. You showed per size yet not per species.
Response 6: Many thanks for pointing this out, and in Figure 9 and lines 448-451 of the article, we have added graphs comparing the metrics of the baseline model and SAW-YOLO on per species and their interpretation.
Author Response File: Author Response.docx
Reviewer 2 Report
Comments and Suggestions for Authors1. In line 171, regarding bottleneck problem, the authors shall specifically explain the deteriorating performance metrics (Equations 9 to 12) and how much the improvement was achieved using the proposed SPD.
2. In line 174, the authors shall provide a rationale as to why four SPDs were selected as the optimal number.
3. The SPD-module, BRA mechanism, DBB, and Wise-IoU proposed in the text refer to existing [31], [34], [35], and [32], respectively. Authors shall clearly explain the difference between their actual contribution and related work.
Additionally, it is recommended that the explanation of the above references be moved to related work section.
4. In line 114, regarding 'web crawler' for dataset construction, the authors shall clearly explain about tool specifications and collection method.
5. Since the actual size of pests generally varies, authors shall describe the actual insect size (mm) in the collected dataset.
6. It is recommended to slightly improve the resolution of Figures 7 and 8.
Author Response
Comments 1: In line 171, regarding bottleneck problem, the authors shall specifically explain the deteriorating performance metrics (Equations 9 to 12) and how much the improvement was achieved using the proposed SPD.
Response 1: Agreed. We have address the impact of SPD on information bottlenecks in conjunction with Table 3 and Equations 9-12, as modified in lines 220-230 and lines 431-434 of the manuscript.
Comments 2: In line 174, the authors shall provide a rationale as to why four SPDs were selected as the optimal number.
Response 2: Thank you for pointing this out, we did not express it clearly enough. Mainly more number of SPD-Modules cannot be reflected in the metrics and the burden increases, specifically in lines 232-235 of the article.
Comments 3: The SPD-module, BRA mechanism, DBB, and Wise-IoU proposed in the text refer to existing [31], [34], [35], and [32], respectively. Authors shall clearly explain the difference between their actual contribution and related work. Additionally, it is recommended that the explanation of the above references be moved to related work section.
Response 3: Thank you for your comments. We add in lines 116-131 of the article the multiple methods we cite and why we cite these structures as part of our related work. More deeply, in lines 272-274 of the article, we describe the thinking behind the use of DBB, and in lines 439-446 of the article, we describe WIOU's contribution to our work.
Comments 4: In line 114, regarding 'web crawler' for dataset construction, the authors shall clearly explain about tool specifications and collection method.
Response 4: Thanks for pointing this out. We add that we use a crawler built in python to crawl on public resources, specifically in lines 157-158 of the article.
Comments 5: Since the actual size of pests generally varies, authors shall describe the actual insect size (mm) in the collected dataset.
Response 5: Agreed. This gives the reader a more complete reference. The actual average diameter of the pest was added to Table 1 of the article.
Comments 6: It is recommended to slightly improve the resolution of Figures 7 and 8.
Response 6: Thanks for pointing this out, Figures 7 and 8 really aren't clear enough due to the amount of information, we have adjusted and enlarged the images as much as possible.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsThe authors faithfully reflected reviewer's comments and improved the quality of the manuscript.