Automatic Water Body Extraction from SAR Images Based on MADF-Net
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis paper has significant application values of water detection from SAR images. The proposed new network framework, MADF-Net, is able to solve the problems of weak extraction ability for detailed feature and low robustness to different water types, which can perform automatic extraction of water areas with high-precision in different water types. The present ideas are interesting. Considering the application values and the contributions of this paper, I would like to recommend that the paper should be accepted for publication after a minor revision.
Below are some comments:
1） In Line 74-134, I suggest the author improve the logical expression of the third paragraph of the introduction, which may help the readers better understand the paper.
2） In Line 152, Please explain why these months are selected as the experimental data.
3） In Line 197-235, in the methodology section, I suggest the expression of deep separable convolution and atrous convolution be more concisely, which can make the paper more compact and easier to understand.
4） In Line 345, please explain what are the meaning of the yellow rectangular boxes in the experimental result diagram.
5） In Line 361, for Table 1 to Table 4, the number of decimal places is not uniform, we suggest keeping two decimal places. In addition, in these tables, maybe % is omitted for PA and IoU, please add it.
Comments on the Quality of English LanguageN/A
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Comments and Suggestions for AuthorsIn the manuscript, the authors proposed the MADFNet: Multi scale Attention Detailed Feature extraction Network to solve the problem of poor detection accuracy of fine tributaries and small water areas in SAR images. In general, the manuscript is well structured and the experiments are abundant. However, appropriate modifications are necessary, and I suggest making minor revisions to this manuscript to further improve its quality. Below are some comments:
1） The revised manuscript seems somewhat confusing in certain parts, highlighting added content prominently but failing to retain track changes for deleted content. Please make further modifications and corrections.
2） In Page 9, Line 320, to my knowledge, Pytorch does not have version 1.20, please check it.
3） In Page 9, Line 333, the network name in reference [19] is SPT-UNet, not MFAM-Net. Please check it. “The tiny tributaries in them are much smaller in size than the main tributaries.”, this sentence is too abrupt and confusing to put here.
4） It is recommended that the author can use different colors to represent missed detections and false alarms in the result, which will enhance clarity.
5） More metrics are suggested for comprehensively evaluate the performance of the network, e.g. time comparison, Params, FLOPs.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
Comments and Suggestions for AuthorsIn this paper, the MADF-Net method is proposed to extract hydrological elements from SAR image data, which is innovative. However, there is still a lot of room for improvement in the focus of the article, the details of the experimental method, and the rigor and logic of the paper.
(1) The language tense of the article writing, the clarity of the pictures and the overall layout of the article need to be further optimized, especially the information of the revision mode should not be retained in the PDF version.
(2) The paper first puts forward the problems of current research on medium and small-scale water extraction and parameter optimization training of extraction model in the introduction, and then puts forward the improved model in the part of research content and methods of this paper, which should especially strengthen the connection between the method in this paper and the problems concerned in this paper and the logical detail description.
- What is the problem concerned in this article and what are the solutions, which need to be further elaborated.
- What is the connection between MADD-Net and the research question and content of this paper? Is the solution relevant and logical to the problem?
- Why not use the current architecture instead of the Unet model that has been mentioned many times in the research review? What are the advantages and disadvantages of the Unet model for addressing the issues of concern in this article? What are the advantages and disadvantages of the current network architecture to address the concerns of this article?
(3) The literature review in section 1 needs to be strengthened.
- First of all, it is necessary to make a clear and detailed summary of the current research status and existing problems, including not only the Unet method, but also strengthen the summary and condensation of other methods;
- Secondly, the literature review needs to strengthen the close connection with the research methods and contents of this paper, especially to summarize the ideas and advantages and disadvantages of each method for solving the problem.
- Is there any connection between the network MFAF-Net mentioned in reference [16] and the MFAM-Net method involved in model comparison in Section 4.2, and the relevant literature is not indicated? What are the differences between MFAF-Net/ FAM-NET and the methods in this article? Are there any similarities? Do we need to run a comparison test? Please complete these details.
(4) The SAR data set produced by the author is mentioned in Section 2 "Materials". Please introduce more details of the data set in detail, including data text introduction and corresponding picture description.
(5) In the first part of the paper, many scholars have carried out related research by improving Unet, but why did DeepLabV3+, MFAM-Net, and GCN be used for reference comparison in the part of model comparison and analysis? Why is there a lack of Unet-based improved models? It should also be compared with the latest SAR water extraction model.
(6) In section 4.2 of the text, the importance and particularity of the three river scenarios selected should be further elaborated.
(7) Why is DeepLabV3+ used as a baseline in Section 4.4?
(8) Section 5 of the article needs to be rewritten, and the models and improvement methods proposed by other scholars should be fully discussed and analyzed.
Comments on the Quality of English LanguageIn the writing of the introduction, methodology, etc., there are many problems such as unclear subject-verb structure and confusing sentence tense, especially the information of revision mode in the pdf version, which makes it difficult to understand the core argument and paragraph logic in the reading process. The structure of the core chapters also needs to be improved. It is suggested that the author should strengthen the refinement of sentence structure and the combing of the logical order of sentences.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 4 Report
Comments and Suggestions for AuthorsIn this paper, a Multiscale Attention Detailed Feature Extraction Network (MADF-Net) is proposed to solve some challenges and problems in SAR image water extraction. High-precision automatic detection of waters of different scales is realized. However, before publication, there are some minor question and comments as follow:
1. The template suggests that the Abstract should be written in a single paragraph of about 200 words maximum. It is recommended to delete unnecessary expressions in the abstract (such as the development of the experiment) to make the abstract more concise and organized.
2. The last paragraph of the Introduction is best to add a summary of the main chapters and contents of the article.
3. Please put the order numbers (a), (b), etc. of the subgraphs in Figure 4, Figure 5, and Figure 6 directly below the subgraphs.
4. Adjust the Figures and Tables to the right position to reduce the large blank in the article.
5. Please modify the Reference format according to the template requirements.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 5 Report
Comments and Suggestions for AuthorsThe comments is attached
Comments for author File: 
Comments.pdf
No comments on Quality of English
Author Response
Please see the attachment.
Author Response File: Author Response.docx
