Enhanced Detection Method for Small and Occluded Targets in Large-Scene Synthetic Aperture Radar Images
Round 1
Reviewer 1 Report
Comments and Suggestions for Authors
-Please insert a Table for introducing your parameters/variables in the very beginning of paper.
-Please compare your work with state-of-art in a tabular format in a new section which is named "Related Work".
-Please give the computational complexity of the proposed method in a mathematical format and compare it with similar work.
-The quality of figure related to Algorithm 1 must be improved.
-It seems the title of paper has a typo "OccludedT"? please describe or correct it.
-
Comments on the Quality of English Language
Please re-check for any typo especially in the title.
Author Response
Thank you for your comments and suggestions. Please see the attachment for details of the modification.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for Authors
This work proposed an occluded or small targets detection method from large scene SAR images exploiting transformer self-attention mechanism into the backbone network based on a GIoU-based loss function. The experimental results demonstrate that the proposed model can automatically recognize and detect small targets in SAR images with significantly high accuracy under various scenarios. The results are convincing and interesting.
The SAR images are generally affected by the speckle noise. Is this proposed method robust with the different levels of speckle noise? The authors should add some results for speckle-noise level versus accuracy. Also, is it also possible to remove the speckle before applying the detection model?
Finally, the authors should discuss recent quantum mechanics-based despeckling methods in this domain, which can also be applied to SAR image processing.
(1) A novel image denoising algorithm using concepts of quantum many-body theory (De-QuIP)
(2) Quantum mechanics-based signal and image representation: Application to denoising (QAB)
Author Response
Thank you for your comments and suggestions. Please see the attachment for details of modification.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for Authors
The paper presents a study of using various attention mechanisms to the problem of ship detection in large-scene SAR images where targets are often small or occluded. The proposed multi-attention mechanism (TAC_CSAC_Net) demonstrates superior detection results compared to other published models.
The experimental analysis uses three publicly available SAR datasets containing over 70k ship images. This is an effective dataset for the experiments and is large enough to yield meaningful results. The observed improvements are significant enough to be noteworthy. The results address only the accuracy (precision and recall) and do not address the computational performance of the model, so we are unable to determine if there is a trade-off required to achieve the improved results and how large it is.
The paper is well-written and clearly presented, with only a few minor typographical errors and suggested re-wording, as noted in the attached.
Comments for author File: Comments.pdf
Author Response
Thank you for your comments and suggestions. Please see the attachment for details of the modification.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for Authors
The authors have revised the paper successfully.
Author Response
We sincerely appreciate your feedback and feel that these revisions will be able to make our paper better and more accurate. Thank you for your patience and guidance.
Reviewer 2 Report
Comments and Suggestions for Authors
The authors addressed all the questions properly and the manuscript is ready for acceptance in my view.
Author Response
We sincerely appreciate your feedback and feel that these revisions will be able to make our paper better and more accurate. Thank you for your patience and guidance.