Wetlands Classification Using Quad-Polarimetric Synthetic Aperture Radar through Convolutional Neural Networks Based on Polarimetric Features

Round 1

Reviewer 1 Report (Previous Reviewer 1)

All previous comments have been well covered.

It is ready for publication.

Author Response

Dear reviewer 1,

The authors, above all, would like to thank you for your comments to help to improve the manuscript. We thank you for taking the time to review the manuscript and appreciate all your comments and suggestions.

Reviewer 2 Report (Previous Reviewer 2)

The authors addressed my comments from the previous version of the manuscript. Further, the manuscript is in much better shape (the abstract, previous and related work, etc.). In general, it is suitable for publication now. There are some typos and minor errors in the manuscript. Thus it can be accepted after a minor revision.

List of typos/errors:

- Section 3.3 Line 380: Exp --> Experimental results

- Line 204-211: please leave blank space before and after enumeration. For example, "features;(2)to examine" --> "features; (2) to examine"

- There are many abbreviations in the manuscript. Please consider to insert a list of abbreviations section in the manuscript. MDPI template offers this.

- "He et al. [5] proposed an efficient-generation countermeasure network" ---> This sentence is rather unclear. Do the authors want to say that He et al. proposed a generative adversarial network?

- Line 103: "convolutional neural networks(CNNs)" --> please leave blank space in these cases: "convolutional neural networks (CNNs)"

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report (Previous Reviewer 3)

The paper seems to be after revisions. In general, the paper shows the use of CNN with polarimetric features for selected classification problem. The idea is interesting, especially in terms of polarimetric features. 

 

The current state of paper is good, but some issues could be extended like:

change the last paragraph in the introduction to the contribution list;

Discuss the latest (from 2022) achievement of remote sensing applications like a combination of roi and deep learning for sonar;

Explain if the polarization decomposition is made on the network. If yes, how.

The caption of Figure 3 is strange, see: "Figure 3. ExperimenFigFigure 3. Experimental flow "

Moreover, fig. 3 should be recreated - maybe some color explaining and more of them. In some blocks, the font is not visible.

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report (Previous Reviewer 4)

This version has been carefully revised, however, there are still some items should be revised.

1.      Keywords should be refined to highlight the creativity of the research method

2.      Some typos should be revised, such as "comparisions"

3.      There are some typographical problems in the paper, such as, line 103- 109, page 3

4.      Some references in the paper are incorrectly formatted and incorrectly cited. Please check. such as [34] and [48]

5.      References should be reviewed and analyzed, not simply listed.

6.      The newly added references [12-14]in the revised paper seem not to be very appropriate, and the latest literature related to the topic should be supplemented, such as

[1]      Land Cover Classification for Polarimetric SAR Images Based on Vision Transformer. Remote Sens. 2022, 14(18), 4656;

[2]      Automatic Design of CNNs via Differentiable Neural Architecture Search for PolSAR Image Classification. IEEE Transactions on Geoscience and Remote Sensing. 2020, 58

[3]      Exploring Vision Transformers for Polarimetric SAR Image Classification, IEEE Transactions on Geoscience and Remote Sensing. 2022,60

[4]      A Deep Reinforcement Learning-Based Framework for PolSAR Imagery Classification. IEEE Transactions on Geoscience and Remote Sensing. 2022, 60

[5]      Complex-Valued Multi-Scale Fully Convolutional Network with Stacked-Dilated Convolution for PolSAR Image Classification. Remote Sens. 2022, 14(15), 3737

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report (Previous Reviewer 3)

The paper was improved according to all my comments. It can be accepted in the current form. 

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

please see attached file!

Comments for author File: Comments.pdf

Reviewer 2 Report

The manuscript seems good. However, it is difficult to read and follow at some points. First of all, I suggest to remove those letters from the abstract which denote mathematical symbols or matrices. It is very uncommon in scientific papers. Further, the abstract is rather weak. Please read some literature about how to write a good abstract (https://writingcenter.gmu.edu/writing-resources/different-genres/writing-an-abstract , https://writing.wisc.edu/handbook/assignments/writing-an-abstract-for-your-research-paper/) and please polish further the abstract. It leaves a good impression to readers and reviewers if the abstract is satisfactory.

Another problem is that the authors do not review carefully the state-of-the-art. In the introduction section, a separate "Related work" subsection would be welcomed. Wetland classification is a very hot research topic. The review and classification of wetland algorithms would be useful for readers. As already mentioned, it is also a popular research topic with many papers, link: https://dblp.org/search?q=wetland+classification . This why, the authors should mention more related paper in a "Related work" subsection (or something like that).

In Figure 1, subfigure e) is probably not too informative. Probably, RGB is very well-known...

After the sentence in Lines 48-49 (We incorporated this principle to classify PolSAR images through convolutional neural network(CNN) in this study), the authors should mention some recent papers to demonstrate that CNNs are very popular in areas of computer vision, such as visual quality (No-Reference Image Quality Assessment with Convolutional Neural Networks and Decision Fusion, 2022), mammogram classification (Transfer learning from chest X-ray pre-trained convolutional neural network for learning mammogram data, 2018), or brain tumor classification (Brain tumor classification in MRI image using convolutional neural network, 2020).

Although the description of the proposed method is good, its novelty remained unclear for me. At the end of the introduction section, a separate "Contributions" subsection is required to this end. Many research paper use a separate "Contributions" subsection and I think it would be a good solution here also.

Please improve the quality of Figure 3. I think its resolution is low. Further, the fourth yellow box is not ideal since the caption in it cannot be read.

In Eq. 9, please use the greek letter instead of "Kappa".

The experimental results section is not a bad but a comparison to at least one other state-of-the-art method is missing.

Reviewer 3 Report

The paper describes the use of polarimetric features for the segmentation/classification task of radar data. The idea is interesting and worth investigating. However, in my opinion, the paper is in the initial state of research and needs many improvements like:

1) Title of the article is misleading - CNN classification algorithm for wetlands? It cannot be in such a version. Change it to more proper text like Wetlands classification using fully... 

2) The abstract could be rewritten to show more details about novelty. 

3) Why the use of polarimetric features is new? 

4) How did you select the learning transfer tool? AlexNet is not the best and not the latest model.

5) At the end of the introduction, a contribution list should be added. It helps readers to analyze the rest of the paper and the new proposition that are further.

6) The used bibliography should be extended. I propose adding a related works section and analyzing the current state of research. Please, analyze mostly the last 2-3 years. I think that the current bibliography list is outdated. Maybe ROI analysis and deep neural networks for sonar data could be also discussed.

7) Explain in more detail the sense, the definition, and the process of creating polarimetric features. Maybe some more examples would be also needed.

8) What is the meaning of a star operator (for instance in Eq. (4))? Multiplication is a \cdot

9) Some pseudocodes could be valuable

10) What are time/computational complexities?

11) Captions of figures should be self-explaining. It should be corrected

12) In your proposal, four schemes are used - why? why such a number of channels, etc.

13) Experimental section is in a basic state: 

a) add comparison with state of art

b) add comparison with other learning transfer models

c) discuss the impact of polarimetric features on the analyzed topic. Compare it with other approaches in the literature

 

Reviewer 4 Report

This manuscript used reflection symmetry decomposition and CNN to do PolSAR image classification. The topic of this manuscript is very useful in polarimetric SAR application and interpretation. The author should issue the following items:

1. The title is 'CNN Classification Algorithm'. However, this work didn't study CNN algorithms, but is an application research of CNN.

2. If the manuscript falls into the research of Algorithm, the novelty is obviously insufficient. If it is an application research, I think the study region (Fig. 1) is not broad enough, and doesn't reflect any conclusive knowledge, e.g., the significance of wetland classification, how to protect wetlands, etc.

3. Introduction of research background and Reference, are insufficient.  The abstract should be rewritten and organized in order to point out the novelty of this manuscript.

4. The literature could be better covered and more recent papers should be added.