Synergistic Attention for Ship Instance Segmentation in SAR Images

Round 1

Reviewer 1 Report

Dear Authors,

Thank you for facing the segmentation methods designed for optical images. It is a significant problem, and your paper describes the principal methods at the state-of-the-art.

Your method based on synergistic attention is interesting, but it is not clear. Equations 1, 2, 3, and 4 need a complete description.

What is the target of equations 1, 2, 3, and 4? Why do you write the convolutions in this form? What problem do they resolve?

AAM (section 3.3) generates a probability map, but it is unclear how you create it.

The general impression of this paper is that you faced many problems providing too much simplistic description. Would you mind providing the source code you used in this research?

Regards

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The authors present an interesting paper about targets detection and generation by combining global attention nodule, semantic attention module, and anchor attention module. The proposed method was proven to be effective by comparing several methods. But I am mainly concerned with the following issues that should be addressed before publication. My major comments and questions are as follows:

1. In line 17, “ Synthetic aperture radar (SAR) ……” appears for the first time in the abstract for abbreviation, here directly with SAR.
2. In line 29, “Improved” should be changed to “improved”.
3. In line 54, “imaging polarity” is usually expressed as “imaging mechanism”.
4. In section 2.2, A literature review of CNN-based ship detection and instance segmentation is not sufficient and should include some latest literature discussions. Then explain the superiority of your proposed method. Some latest literature is listed as below:

Guo, H.; Yang, X.; Wang, N.; Gao, X. A CenterNet++ model for ship detection in SAR images. Pattern Recognition 2021, 112, 107787.

Geng, X.; Shi, L.; Yang, J.; Li, P.; Zhao, L.; Sun, W.; Zhao, J. Ship Detection and Feature Visualization Analysis Based on Lightweight CNN in VH and VV Polarization Images. Remote Sens. 2021, 13, 1184.

Fu, J.M.; Sun, X.; Wang, Z.R.; Fu, K. An Anchor-Free Method Based on Feature Balancing and Refinement Network for Multiscale Ship Detection in SAR Images. IEEE Trans. Geosci. Remote Sensing 2021, 59, 1331-1344, doi:10.1109/tgrs.2020.3005151.

Wu, Z., Hou, B., Ren, B., Ren, Z., Wang, S., & Jiao, L. (2021). A deep detection network based on interaction of instance segmentation and object detection for SAR images. Remote Sensing, 13(13), 2582.

de Albuquerque, A. O., de Carvalho, O. L. F., e Silva, C. R., de Bem, P. P., Gomes, R. A. T., Borges, D. L., ... & de Carvalho Júnior, O. A. (2021). Instance segmentation of center pivot irrigation systems using multi-temporal SENTINEL-1 SAR images. Remote Sensing Applications: Society and Environment, 23, 100537.

5. I propose to adjust parts 2.3, 2.4, and 2.5 to 2.1, 2.2, and 2.3 before discussing ship detection based on SAR image.
6. In line 108, “synergistic” instead of “Synergistic”.
7. In lines 233-237, the statement about “ The traditional residual block uses convolution (Conv) with kernel size of 11 to reduce and expand the feature map dimension, so that the number of filters of the convolution with kernel size of 3×3 is not affected by the input of the previous layer, ……” is confused. In the residual block, 1×1 convolution was used to reduce and expand the feature map dimension.
8. The variable x should explain in the first occurrence.
9. In the experiments section, the method proposed by the authors is validated by comparing it with other methods but neglected efficiency. The efficiency of detection or instance segmentation compared to other methods should explain. Is the efficiency still better than other methods while maintaining high accuracy detection?
10. In section 4.2, the initialized anchor size by cluster analysis should be added.
11. I suggest the authors should add a discussion section on the advantages and disadvantages of the model compared to other methods.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The article deals with the issue of a segmentation methods designed for optical images.

The topics covered in the work are interesting and needed. The article contains the state of literature knowledge in the subject of paper. However, there are no literature references to mathematical formulas.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Dear Authors,
Thank you for replying to all my answers.

Regards