Marine Oil Spill Detection from SAR Images Based on Attention U-Net Model Using Polarimetric and Wind Speed Information
Round 1
Reviewer 1 Report
Oil spills can exert damages to the health of the marine environment. This is an interesting and well-written paper which study the topic of oil spill detection using PolSAR images based on deep learning method. My detailed comments are as follows:
1) Generally speaking, the related papers are cited properly and the research situation are well introduced, but more references about SAR image classification or target detection published in the last three years can be introduced to help the readers better understand the research situation.
2) The authors claimed that, by considering the wind speed information, the detection accuracy of the proposed method can be improved. Please consider to validate the above assumption using some more experiments and figures.
3) The format of the reference list should be standardized.
4) Are polarimetric decomposition parameters and the wind speed also employed into the U-Net model used for comparison in the experimental part?
Author Response
Please refer to the attached file.
Author Response File: Author Response.doc
Reviewer 2 Report
This paper proposes a new model for PolSAR oil spill detection, in which an attention U-Net architecture is employed to ensure the model can focuse more on feature extraction and better distinguish oil films from look-alikes; more importantly, both the rich polarimetric information of PolSAR data and the wind speed information of the sea surface are utilized when training the model.
A few questions are for reference only:
Question1:
Figure2 shows the architecture of the AUOSD model. How do you integrate wind speed information into the model?
Question2:
In Table3 we can see different methods have different result. But different hyper parameters have an impact on different methods. Are the results listed in Table 3 the best hyper parameter settings for the corresponding method?
Question3:
Is there any evidence that red tides are dark spots? Low wind area, bio-oil film, upwelling, back mountain wind, ship wake, atmospheric gravity wave, rain packet, internal wave, etc. show dark spots on the SAR image.
Abstract and conclusion are the essence of the whole article, but the abstract and conclusion do not accurately reflect the conclusion of the whole article, so it is suggested to be revised.
Author Response
Please refer to the attached file.
Author Response File: Author Response.doc
Reviewer 3 Report
I reviewed the paper titled “Marine Oil Spill Detection from SAR Images Based on 2 Attention U-Net Model Using Polarimetric and Wind Speed 3 Information”. Comments on the manuscript entitled are listed below.
1. Lines [85-90], the sentence is too long.
2. Line [96], correct the section title to be “material and methods” only.
3. Line [98], there must be some introduction about the subject, e.g., PolSAR Oil Spill Detection Dataset, before starting expressing “ In this study, we used”. A brief explanation of the subject, types, importance, must be added first.
4. More information about the used location must be added, where they are located, their physical description, and why only these locations were implemented in this study. e.g., are not there other similar locations?
5. The discretion of the developed technique is short and little confusing.
6. Section 3.1 must be moved into material and methods.
7. What are the ranges of metrics of equations 7-10? And what their outcomes mean?
8. Section 4 must be re-written. As it is now, it is more like an abstract.
Author Response
Please refer to the attached file.
Author Response File: Author Response.doc
Round 2
Reviewer 3 Report
i still believe that the conclusion needs to be re-written. added new sentence would not improve the content, it stills like an abstract of the paper.
Author Response
Dear reviewer, thanks very much for your constructive suggestion. In this newly revised paper, we have re-written the conclusion section, including, introduced more conclusions which were found in the experimental results, discussed the drawbacks of this work, and described what we will do in our future work.