A General Image Super-Resolution Reconstruction Technique for Walnut Object Detection Model

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

 

1. The research process in Figure 1 is too simplistic, lacks crucial information, and fails to highlight distinctive features.
2. Figure 2 needs optimization to emphasize key points.
3. With the addition of super-resolution reconstruction, does the processing efficiency decrease significantly enough to affect its practicality? It is recommended to compare the differences in processing time.
4. It is suggested to provide intuitive figures to demonstrate the improvement in accuracy.
5. From Table 2, it appears that the improvement in Precision (P) is more significant, while the improvement in Recall (R) is relatively small. What is the reason for this? Please analyze this in conjunction with the corresponding extraction result figures.
6. How well does this algorithm adapt to data collected at different times and weather conditions? For instance, would images of walnut fruits at different stages of maturity affect the model's performance? Images collected under different weather (lighting) conditions may have significant quality variations. Can the trained model be applied to images collected on another day? In particular, the effect of super-resolution reconstruction may be influenced by image quality and the contrast between targets and background.
7. The experimental results should include some specific comparison figures of detection results.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

I have the following concern.

1. It is not clear what the input resolution of the images was and what it became after using the MDAARB, RRDB, CBAM modules. Show it step by step in pilseles.

2. To assess the quality of image processing, two PSNR, SSIM metrics are not enough. It was necessary to use DS-SSIM, NIQE metrics.

3. It is known that many iterations are required to increase the resolution of walnut images. In your case, it is 500,000 iterations, which increases the processing time. Therefore, in Tables 2, 3, 4, you need to provide a comparative estimate based on the FLOPS parameter.

4. Equation (7)-(9) are standard and known in ML, so they can be eliminated.

5. And where are the comparative quantitative estimates with other models.

6. The Discussion section is missing. In addition, you need to more clearly show the limitations of your model and the perspective of further research.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

1. As I said before, given that you have 2490 images, you should give more detailed images of the experimental results to show how the super-resolution reconstruction results improve the recognition. The cases given so far are too few for me to judge the effectiveness of the method.
2. Figure 1 still needs to be optimized, and you should reflect that your research focuses on improving recognition accuracy through super-resolution reconstruction.
3. Figure 2 needs to be redrawn, you should put information related to the research content, such as satellite image maps of the study area, UAV flight trajectories, examples of images collected at different points, etc. The current DEM and NDVI are not related to the research content.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

I am satisfied with the answers to my concerns. The changes and additions made have greatly improved the perception of the results obtained.

Author Response

Thank you again for your valuable feedback. Your comments and suggestions have significantly improved our manuscript.