Oil Spill Detection by CP SAR Based on the Power Entropy Decomposition
Round 1
Reviewer 1 Report
This paper aims at transferring the Lesa parameter based on entropy decomposition theory in ship detection to SAR oil spill detection, and compact polarimetric SAR data is used to compare the Lesa parameter and the other five characteristics. And through the random forest and variable importance analysis, the advantages of Lesa parameter in oil spill detection are demonstrated.
The following issues need to be addressed by the authors:
1. Page 1-Line 44: More experiments on the comparative results of full polarimetric (FP) SAR and compact polarimetric (CP) SAR should be added to the introduction part of past research to prove the concept of "compact polarization is better than full polarization" mentioned in this paper.
2. The three scenarios mentioned in Figure 1 should be briefly explained in the paragraph before the picture.
3. Line 215: Explain the basis for the division of J-M distance (two critical points of 1 and 1.9), if there are references, please indicate them.
4. For the superiority of Lesa in SAR oil spill detection, this paper adopts the variable importance in random forest for ranking evaluation. It is recommended to use several machine learning algorithms to enhance persuasiveness by analyzing and comparing the results of different algorithms.
5. On what criteria are the remaining five features selected for comparison? If there are more suitable characteristics than them, please provide an additional explanation.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 2 Report
I think the work is worthy of publication to promote discussion concerning this attempt to provide a new tool for oil spill detection. Though, the text need improvements (e.g., correction of sentences, typo erros).
Below are other suggestions to boost the work.
ABSTRACT
Line 16-17: Relevant quantitative results should be included in the abstract (you wrote: We compare ???? with the 16 other five popular polarimetric features and validate it by quantitative evaluation.)
The abstract must includes at least Gap & objective, Method, Results, and Conclusion. Consider to include a more information about Method in abstract.
INTRODUCTION
Line 27: "At the same time, oil spills caused by 27 the massive exploitation of oil have become a concern for people." Concern for people? Re-write the sentence because it is an issue for environmental conservation.
METHOD
Try to improve this section including a contextualization of the methodology. Read previous works on the subject. Starting the section 2.1 with an equation is not the right way.
CONCLUSION
Go straight tp the point and include quantitative indicators in your conclusion to make it more useful and based on the results obtained .
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
The manuscript 'Oil spill detection by CP SAR based on the power entropy decomposition' introduces Power-Entropy decomposition theory to CP SAR oil spill detection and utilizes a low-entropy radiation amplitude parameter as the polarization feature. There are some comments and suggestions as follows:
(1) In Fig.2, the horizontal and vertical coordinates should be geographical coordinates.
(2) The full name of 'SPAN' should be given, and the equation of variable importance ranking should be formulated in the manuscript.
(3) For the description 'Compared with the results of manual segmentation, the ...', the results of manual segmentation should be compared with the classification results of oil slicks in the manuscript. In addition, the manual segmentation standard for distinguishing thin and thick oil slicks should be introduced in the manuscript.
(4) Oil spill classification based on random forest is a traditional method. With the development of deep learning, oil spill detection methods based on CNNs have been studied widely such as 'Oil spill detection with multiscale conditional adversarial networks with small-data training' and 'Oil spill identification from satellite images using deep neural networks'. These CNN-based detection methods play an important role in the oil spill detection area. Therefore, it is necessary to present this kind of methods in the Introduction.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
The author has made careful revisions to the problems involving the experimental process and format errors in the article.
Regarding the fourth point in the comments, I would like to make further supplementary explanations here. For the Lesa feature proposed in the article, the results of random forest prove its superiority, but is this superiority a special case, that is, when other machine learning methods are used to verify, will the same conclusion as the article be obtained? Further experiments by the authors are required.
Author Response
Please see the attachment.
Author Response File: Author Response.docx