Deep Multi-Scale Recurrent Network for Synthetic Aperture Radar Images Despeckling
Round 1
Reviewer 1 Report
The paper includes a comprehensive review of state of the art regarding speckle noise and SAR despeckling techniques.
An end-to-end network multi-scale recurrent (MSR) network- for SAR image despeckling is proposed. The network can automatically perform despeckling instead of using additional division operation or subtraction operation. MSR-net is built based on cascaded subnetworks buts share the same training parameters in the net.
The number of parameters that need to be trained seems to be notably decreased with respect other in the literature.
Experimental settings and proposed evaluation criteria (capable of evaluating the edge and typical features retention ability in despeckling tasks directly) seems to be adequate for assessing results of the proposed method, and in particular the quality assessment of reconstructed images.
Synthetic and experimental results are compared with other existing despeckling techniques: performance of our MSR-net with other three methods, SAR-BM3D, and RED-Net, which confirms adequate assessment methodology. High scores of the proposed method results in such comparison show the relevance of such proposed method.
The paper is well written and well structured ending with an appropriate discussion.
Regarding formatting just revise:
Fig.6 caption (convolution(deconvolution).))
Author Response
We feel great thanks for your professional review on our article. According to your advice, we modified the caption of this figure in the revision. Please check it on figure 6 in the paper.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The paper proposes a deep learning solution for SAR images Despeckling. The paper is well structured, however, the details of the CNN and LSTM and other performance evaluation parameters need to be described more concise. Specifically, the information given in Sections 2 and 3 can be summarized and explained in another language since some sentences seem to be used from other references without citing or change in the sentences. Moreover, the present form is too long compared to the main section of the paper (sections 3 and 4). Please try to elaborate more on the proposed method and the results.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The Introduction section is well-written and comprehensive. The proposed idea is slightly novel. The Simulation section could be improved. More specifically,
- Section 4, the results in Table 2-4 should be explicitly explained. The authors are suggested to provide a short discussion on why we have different performances for the building (SSIM, RED-Net) in Table 2, and for the freeway image (EFKR, RED-Net) in Table 3, and for the freeway & airplane in Table 4.
- Section 5, the choice of scale: The authors are expected to discuss the general rule for choosing "s". They are suggested to do perform simulations on not only one or two images, but rather on a fair set of images and come up with an experimental value for selecting "s".
Minor editorial concern:
-Conclusion section, 1st paragraph: "can to" should be "can".
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
I had the opportunity to review the "Deep Multi-scale Recurrent Network for SAR images Despeckling" manuscript. The manuscript has scientific merit and is within the scope of Remote Sensing. However, I have major concerns before considering it for publication. Below I put some comments for the authors:
- the manuscript is not in accordance with the model provided by Remote Sensing;
- I suggest removing the term SAR from the title and writing it in full;
- Authors need to put more results in Abstract. In addition, it is necessary to end it with a sentence demonstrating the importance of this study;
- I suggest removing Figure 1 from the Introduction;
- There are several excerpts in the Introduction that need literature to support the information presented;
- The Experiments and Results section is confusing. I suggest the authors to separate into two independent sections: one with the experiments and one with the obtained results;
- You need to provide more details about the data set used. Will it be made available to readers?
- You must provide a caption for Tables 1, 2, and 3. What do underlined or bold values ​​indicate?
- The manuscript's conclusions need to be summarized and objectively answer the proposed hypotheses and objectives.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
The responses of the reviewers are satisfactory. Mainly, the response to my earlier comment on the selection of the scale factor is good and they justified it well via the plots when varying S. The choice s=3now seems more reasonable.
Reviewer 4 Report
The authors made all the corrections I requested.
Therefore, I recommend the manuscript for publication.
