An Improved S²A-Net Algorithm for Ship Object Detection in Optical Remote Sensing Images

Round 1

Reviewer 1 Report

The Research Paper needs the following revisions and is subject to re-review. After re-review the final decision for the paper will be made:

Comments:

1.     The application of D4 should not be one of the contributions.

2.     Figure 8 and Figure 9 have the same title.

3.     The explanation of the parameters after the equation should be standard.

4.     Why only use mAP as a metric? There are many other metrics such as accuracy, recall, F-score, ROC, and AUC, which can better reflect the performance of the proposed method.

5.     “image weights sampling” in Table 3 should be in upper case.

6.     The lack of subtitles for the figures increases the difficulties of understanding.

7.     Figure 13 lacks the information of ground truth, it’s hard to compare the results of different methods.

8.     Some of the discussion for the results can hardly be obtained from the table or image directly, for example, line 390-391 “it is not conducive to extracting sufficient contextual information from a densely distributed ship scene for small convolution kernels”.

9.     Why compare the following kernel sizes in Table 6 rather than other combinations of kernel sizes?

In sum, I think this work is meaningful, but the contribution seems to be not outstanding for Remote Sensing.

Author Response

Thank you very much for your careful review and constructive suggestions with regard to our manuscript. These comments are very helpful for us to revise and improve our paper. We have studied these comments carefully and tried our best to revise and improve the manuscript. Please see the attachment. 

Author Response File: Author Response.pdf

Reviewer 2 Report

I had the pleasure of reviewing the above article. The article can be accepted if the following minor revisions are implemented:

To date, S2A-Net has not been widely used in the scientific community. Explain briefly in your article why you chose S2A-Net for your experiments?

Figure 3. was taken from another source. Either specify the source in the caption or create your own image. This applies to all images from external sources.

Briefly disscuss in your article why the class ship subclass results when compared to common deep learning models in Table 9 do not match the model comparisons of the authors of the FAIR1M data article (FAIR1M:A Benchmark Dataset for Fine-grained Object Recognition in High-Resolution Remote Sensing Imagery , see also Large Selective Kernel Network for Remote Sensing Object Detection), e.g., for the Cascade R-CNN and Gliding Vertex networks for e.g., class pasanger ship? In both articles Cascade R-CNN performs better than Gliding Vertex for class Ps and not in your comparison. 

Double check again your English sentence structure and spelling in your article.

Author Response

Thank you very much for your careful review and constructive suggestions with regard to our manuscript. These comments are very helpful for us to revise and improve our paper. We have studied these comments carefully and tried our best to revise and improve the manuscript. Please see the attachment. 

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors have presented an improved algorithm for object detection in optical remote sensing images. The (S2A-net) is one of the well-defined algorithms for various applications. Authors have compared the improved version of the Improved S2A with previous S2A & obtained some better outcomes on the basis of three evaluation matrices i.e., Precision, Recall, AP & mAP. However, some of the major observations are as follows.

1.     L16, S2A-net needs to be specified properly. Mention its full form, and are characteristics and advantages/disadvantages of the original S2A.

2.     L35, revise it as “Compared with infrared remote sensing images and synthetic aperture radar (SAR) remote sensing images”

3.     L40, “etc.??” mention any other if any category is there otherwise remove etc.

4.     I have observed many abbreviations throughout the manuscript are not specified properly. Such as S2A, YOLO, GPU, RCNN etc. Please recheck.

5.     Table 1, defines all the parameters in footnotes utilized in this table. Check the same for other tables.

6.     Figure 3 (S2A Structure), Could you highlight (with proper mention) the improved points with respect to the original S2A (If possible)? However, the authors have explained the same step in subsequent sections.

7.     Sub-numbering in the Figure 11 & 12 is missing. Check the same for other tables.

8.     Table 3, what is the meaning of “tick”? mention in footnotes. Check the same for other tables.

9.     No need to define any abbreviation in the conclusion

10.  The reference section could be improved with more number of recent references.

Author Response

Thank you very much for your careful review and constructive suggestions with regard to our manuscript. These comments are very helpful for us to revise and improve our paper. We have studied these comments carefully and tried our best to revise and improve the manuscript. Please see the attachment. 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Dear authors, thank you for taking care of my comments. All the reviewers' comments have been addressed. Two more points I may given are as follows:

1. Maybe some of Response 8  can be added to the manuscript for better understanding.

2. There is a wrong cite in Line 464, maybe you should recheck again.

Author Response

Thank you very much for your careful review and constructive suggestions with regard to our manuscript. These comments are very helpful for us to revise and improve our paper. We have studied these comments carefully and tried our best to revise and improve the manuscript. Please see the attachment. 

Author Response File: Author Response.pdf

Reviewer 3 Report

All the queries have been addressed by the authors.

Author Response

Thank you very much for your careful review.