Multi-Aspect Convolutional-Transformer Network for SAR Automatic Target Recognition

Round 1

Reviewer 1 Report

This paper studied the problem of multi-aspect SAR target recognition based on deep learning and proposed a novel method based on convolutional autoencoder and transformer. The paper is well written, and the performance of the proposed method is fully demonstrated by experiments, which is better under various conditions than some existing methods. But still, there are some problems to be further improved before its publication. The following problems are numbered for convenience.

1. More relevant research background can be supplemented in Introduction, especially the recent SAR target recognition methods.

2. The difference between your proposed method and the existing method needs to be highlighted in the paper.

3. It is mentioned in the paper that for feature extraction the parameters of the first two layers of the CAE’s encoder after training are frozen but the parameters of the last layer are fine-tuned while the overall network training. You need to explain this design in detail.

4. In your experiment, the angle range when constructing the multi-aspect sequences is set to 45°, which also needs to be explained.

5. There is at least one format error in the manuscript, such as, in page 12, TABLE 2, the first letter of the word “total” needs to be capitalized. Please check the manuscript carefully.

6. The discussion on the backbone of the proposed network is not complete. In this paper, only the results of different network structures for feature extraction are given in table 14. It is better to give the results and analysis with different parameter settings for the transformer encoder.

Overall, it is a nice work and is possible to be published after fully consideration of aforementioned issues.

Author Response

Dear Reviewer:

Thank you very much for the time and effort that you have put into reviewing the previous version of our manuscript entitled “Multi-aspect Convolutional-Transformer Network for SAR Automatic Target Recognition” (remotesensing-1848218).

There is no doubt that these comments are valuable and very helpful for revising and improving our manuscript. We have carefully considered every comment and made some corrections to the original manuscript. Please see the attachment for details of revision.

Best regards!

Siyuan Li, Zongxu Pan, and Yuxin Hu

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper proposed a multi-aspect SAR recognition method based on self-attention, which is used to find the correlation between the semantic information of images.

Here are my concerns:

1.       Avoid lumping references as in [x-y], [x, y] and all other. It is not necessary to give several references that say exactly the same.

2.       Avoid using first person.

3.       The overall layout needs to be rechecked. Paragraphs should not be separated by tables or figures.

4.       The author should recheck the full text. There are many detail errors. Such as the garbled code at the bottom of the first page.

5.       The results of your comparative study should be discussed in-depth and with more insightful comments on the behaviour of your algorithm on various case studies. Discussing results should not mean reading out the tables and figures once again.

6.       How is the performance of the proposed algorithm compared with the following algorithms: “SAR Targets Classification Based on Deep Memory Convolution Neural Networks and Transfer Parameters”, “Dense connection and depthwise separable convolution based CNN for polarimetric SAR image classification”.

7.       The authors should make sure the conclusions reflect on the strengths and weaknesses of their work, how others in the field can benefit from it and thoroughly discus future work.

8.       According to the Reference style, the title format should be uniform.

Author Response

Dear Reviewer:

Thank you very much for the time and effort that you have put into reviewing the previous version of our manuscript entitled “Multi-aspect Convolutional-Transformer Network for SAR Automatic Target Recognition” (remotesensing-1848218).

There is no doubt that these comments are valuable and very helpful for revising and improving our manuscript. We have carefully considered every comment and made some corrections to the original manuscript. Please see the attachment for details of revision.

Best regards!

Siyuan Li, Zongxu Pan, and Yuxin Hu

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

It can be accepted.