PolSAR Image Classification Based on Multi-Modal Contrastive Fully Convolutional Network
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsPlease find the attached file.
Comments for author File: 
Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsIn this paper, a deep learning method is proposed for the field of PolSAR image classification which has higher classification accuracy. Some specific comments are given below.
1. The author should add more recent deep learning based PolSAR image classification methods in the article introduction.
2. In the experimental design of Section 3.2, whether multi-modal features are applied during the experiments of CL-CNN and CL-FCN, please explain how the positive and negative samples are constructed during the contrastive learning process of these two sets of experiments.
3. In page 14, Plots (c1) and (d1) in Figure 10 do not match images (c) and (d), please check for editing errors.
4. What is the difference between the fully convolutional network proposed in this article and traditional fully convolutional networks? What are the advantages?
5. The main contribution of this work lies in the development of a multi-modal contrastive FCN model. What are the advantages of this model compared to convolutional based contrastive learning method? What is the difference between the selection of positive and negative samples in this model and the selection of positive and negative samples in convolutional based contrastive learning method?
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThis paper proposes a multi-modal contrastive fully convolutional network for PolSAR image classification. This is an interesting and significant topic to improve the accuracy of PolSAR image classification based on deep learning in small samples. The experimental results confirm the performance of this method. However, there are some unclear parts in the manuscript.
1. The description of the proposed method in the abstract section is too simplistic. Please reorganize and introduce the content of this section!
2. If the introduction only appears once in the paper, it does not need to be abbreviated, such as Page 2, line 47, The abbreviation for decision tree (DT) only appears once.
3. Compared with the traditional pixel-based contrast learning method, the convolutional network adopted also considers the spatial relationship between pixels. What are the advantages of the proposed method compared with the traditional pixel-based contrast learning method? Is the computational complexity higher?
4. The experimental part lacks a comparison with the latest PolSAR classification methods, so it is suggested to add one or two new PolSAR image classification methods based on deep learning.
5. Figure 9 is not readable enough, and special notations should be added at all key points to improve understanding of the curve changes in Figure 9.
6. In Eq.6, what does variable E represent, and what does variable z represent in Eq.7?
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsI have no more questions.
