Multi-Featured Sea Ice Classification with SAR Image Based on Convolutional Neural Network

Round 1

Reviewer 1 Report

In this paper, a multi-featured synthetic aperture radar (SAR) method is proposed. By combining polarization features calculated by polarization decomposition and spectrogram features calculated by joint time-frequency analysis (JTFA), the proposed method can achieve higher accuracy with smaller data volume and computational effort than the single-feature sea ice classification method. Specifically, the reviewer has the following comments on this work:

 

1.         Pauli decomposition is used to extract the polarization features of sea ice. Are there any other methods that can extract the polarization features of sea ice, and is there any specific reason for utilizing Pauli decomposition?

2.         As shown in Figure 12, different sizes of patch cause different trends, the reason why the size can affect the trend should be explained.

3.         You should give a detailed explanation of the difference between producer accuracy, user accuracy precision and recall in the manuscript.

4.         The innovations and contributions of this paper are not clear enough. It is necessary to specify them in the abstract and introduction.

5.         The authors tried to conclude the recent advances. However, there are still some relevant works about the application of deep learning-based technology in SAR. It is recommended to add these works to the reference, such as

https://www.doi.org/10.1109/LGRS.2020.3018186

https://www.doi.org/10.1109/TGRS.2023.3248040

https://www.doi.org/10.1109/JSTARS.2018.2871556

 

Minor editing of English language required

Author Response

Thank you very much for the time and effort you and the reviewers put into reviewing our articles. We are truly grateful to your and other reviewer’s valuable comments and thoughtful suggestions on our manuscript titled “Multi-featured SAR sea ice classification based on convolutional neural network”. Based on these comments, we have made careful modifications on the original manuscript. These comments have greatly helped us improve the quality of the manuscript. We have prepared two versions for the new submission, an original version and a highlighted version, where all changes made to the manuscript are highlighted in yellow color. We hope that these revisions are satisfactory and the revised version will meet your standard. The point-by-point response to the comments of editor and reviewers can be found below.

Author Response File: Author Response.pdf

Reviewer 2 Report

Thanks for the authors to submit the article “Multi-featured sea ice classification with SAR image based on convolutional neural network”. After reading this paper, I think it is an effective paper, where the introduction, methodology and the figure/tables are presented clearly. I can feel author’s efforts and serious attitude for this paper. Though the novelties of this paper (application of multi-features) are limited, I think it still is an valuable research work. But it needs to be major revision for further improvement before publication. The comments are listed as follows.

 

1. The Pauli-decomposed features have been applied in many image interpretation cases, and the joint time-frequency analysis also has been applied in the existing researches as mentioned in the paper. So that these features are not novel, why you still select them to classify sea ice types. I think the importance should be presented in the Abstract and Introduction sections from the perspective of sea ice classification. For example, the importance of the pauli-decomposed intensities or polarimetric phases for the sea ice classification should be described.

The novelties of feature combination are boring for the PolSAR classification, so you should list some novel or interesting insights by yourself to strengthen the technical innovations.

 

2. Furthermore, it is hard to understand the reason of phase features utilization. Sea ice is a kind of distributed objects, which is adoptive to be described using second-order statistics or neighboring average features. But the phase features are presented with single pixel. is it appropriate for sea ice, and can you consider the interference of speckle noise for classification. If you use the phase information, the filtering can not be utilized. Please explain it.

 

3. Why do you select the Pauli-decomposed features. The Freeman-Durden decomposition, Yamaguchi Four-component decomposition method have been applied in the sea ice classification. Why they are not be selected?

 

4. The obvious shortcoming of this paper is the results and discussions. As for the result analysis, as shown in Table 6, your accuracy with combined features and the accuracies with single kind of feature have obvious difference. This is hard to understand, the combination of two kinds of bad features (70% and 80%) can achieve very good result (>90%). Please seriously explain it and add the corresponding analysis.

 

5. As for the discussions, I think you should add the comparison with your methods with other algorithms with two kinds of features, or add the comparison with your utilized features with other common features. Otherwise, it is hard to prove the effectiveness of your method. It should be revised.

 

 

6. Line 118-124: the first and third point can not be presented as the novelties, please remove them, and mainly describe the technical novelties of your method. These can be described in the manuscript as details.

 

7. Section 4.0.X of section serial number is inappropriate, these can be defined as 4.1.x.

 

8. Line 65: polarized SAR should be the polarimetric SAR.

 

9. Line 68: polarization decomposition should be the polarimetric decomposition.

 

10. Line 60:”And for the ”，And can not be used in the beginning of sentence.

 

You should use official definitions and words for polarimetric SAR classification. and features. 

Author Response

Thank you very much for the time and effort you and the reviewers put into reviewing our articles. We are truly grateful to your and other reviewer’s valuable comments and thoughtful suggestions on our manuscript titled “Multi-featured SAR sea ice classification based on convolutional neural network”. Based on these comments, we have made careful modifications on the originalmanuscript. These comments have greatly helped us improve the quality of the manuscript. We have prepared two versions for the new submission, an original version and a highlighted version, where all changes made to the manuscript are highlighted in yellow color. We hope that these revisions are satisfactory and the revised version will meet your standard. The point-by-point response to the comments of editor and reviewers can be found below.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have revised the manuscript according to the comments of all reviewers. I have no other questions.

Minor editing of English language required.

Author Response

Thank you for recognising our previous revisions and we have checked the English presentation of the article.

Thank you again for your valuable input and time in reviewing our article.

Reviewer 2 Report

Thanks for the careful revisions, I think most of comments have been responsed clearly. I think the technical novelties also have been improved effectively. Most grammar errors have been corrected. 

Before publication, I think there also has a little problem or suggestion for your discussions. In your responses, the pauli-decomposed features are beneficial for the sea ice type discrimination ability, while the JTFA features are good at the edge detection. Your result analysis and discussions just focus on the classification accuracy, which is corresponding to the pauli-decomposed features. So that it is hard to understand the advantage of JTFA feature, of which advantage is not accuracy. Therefore, I suggest that you can add the edge detection accuracy, such as MIoU accuracy, Boundary Recall or Achievable Segmentation Accuracy. These indices can make your comparison about JTFA feature more complete. 

Thanks for your postive attitude, hope your paper can be further improved and it can be published successfully. 

Author Response

Thank you very much for recognising our previous modifications. Our description and discussion on JTFA in favour of edge feature detection is indeed insufficient. Therefore we have added this part of the discussion on page 22, line 582-592. We used the IoU accuracy and illustrated it with the actual classification results according to your suggestion.

 

The classification results obtained by the Pol. Decomp. method showed more accurate class judgment, while the JTFA method provided better description of the morphological edge information of different classes of sea ice, especially in the classification of NI and FI, as shown in Figure 19. In addition, we also performed the calculation of Intersection over Union (IoU) accuracy as a supplementary note. IoU is a commonly used metric in computer vision tasks and is computed with the formula: IoU = (Intersection area) / (Union area). IoU accuracy values range from 0 to 1 and are useful metrics for assessing the quality of object segmentation results. As shown in Tab 7, the effectiveness of the JTFA method in segmenting sea ice is sometimes at an advantage in the classification of NI and FI. The multi-featured method also combines these advantages to improve the accuracy of sea ice classification.

We hope that these revisions address your concerns and improve the overall quality of our manuscript. Thank you again for your feedback and your time in reviewing our work.

 

Author Response File: Author Response.pdf