Deep Multi-Order Spatial–Spectral Residual Feature Extractor for Weak Information Mining in Remote Sensing Imagery

Round 1

Reviewer 1 Report (Previous Reviewer 2)

Comments and Suggestions for Authors

Although most of the comments are answered in a suitable manner to improve the manuscript, some revisions are also needed.

1. It seems that two versions of Figure 3 are presented in the manuscript. The authors should check Figure 3 again.  

2. The explanation and expression of using the self-made Pasture dataset is still insufficient. Although the authors have explained the results of Table 5 and Table 6 in line 436 to line 441, some detailed information is ignored. It can be found that the accuracies for most classes (excluding the “Leymus chinensis” class) are quite low even though the proposed method is used to improve the result. Especially for the “Thalictrum” class, the accuracy is only about 0.1. What is the significance of actual measurement with such low accuracies? Therefore, the authors’ response (this study also want to test the proposed method can used to RGB image or not, and have a not bad performance.) is unconvincing.

3. The authors have stated in the text “RGB image also is the an important data in the remote sensing” (line 216 to line 217). It seems that there is a grammar error.

Author Response

Dear Reviewer,

Thank for your kindness working and valuable advice.

We revised the manuscript carefully according to your comments and suggestions, and provide the point-by-point responses in submiit word file. All changes were marked in red using the “Track Changes” function in the revised manuscript.

Best regards,

Xizhen Zhang

 

Author Response File: Author Response.docx

Reviewer 2 Report (New Reviewer)

Comments and Suggestions for Authors

The paper presents a method for classifications of hyperspectral images based on feature extractions with spatial and spectral contexts. the methodology is described in detail and the experiment results are well organized. I think the paper is acceptable for publication from the journal, but I would leave some comments for a minor revision.

1. I understand the method is innovative in extending FE to spaital contexts. However, the method handles spatial features in one dimension, though image data has two dimensions. Even with the reduced dimension, the feature extraction process may capture proximity of pixels neighboring in a row. However, it indicates a model trained with an image of a size, such as  2517 × 2335 pixels, is not useful to the other images of a different size, such as 1999 194 × 2051 pixels. I think the discussion could be extended for future works on FE process retaining spatial proximity of pixels in rows, so that trained models will be flexible for applications. I am not very sure if my understanding is correct, but would appreciate addressing it.

2. The results with PT dataset may extend discussions on effects of spatial features in classifications. While remote sensing images are with more gradient spatial autocorrelations between pixels, close-range images like PT dataset are more discrete split by objects. Therefore, the method is expected to perform better with coarse-scale spatial resolution. If this insight is correct, please consider adding the above to the discussions.

3. I would suggest adding a reference to a paper below. The paper presents a method to handle spatio-temporal features in deep learning for land cover classification. Although the dimension is much different, the paper approaches similar problems, I think.

Truong, V.T., Hirayama, S., Phan, D.C. et al. JAXA’s new high-resolution land use land cover map for Vietnam using a time-feature convolutional neural network. Sci Rep 14, 3926 (2024). https://doi.org/10.1038/s41598-024-54308-1

Thank you in advance for your consideration.

Author Response

Dear Reviewer,

Thank for your kindness working and valuable advice.

We revised the manuscript carefully according to your comments and suggestions, and provide the point-by-point responses in submiit word file. All changes were marked in red using the “Track Changes” function in the revised manuscript.

Best regards,

Xizhen Zhang

Author Response File: Author Response.docx

Reviewer 3 Report (New Reviewer)

Comments and Suggestions for Authors

The manuscript shows the author’s hard work. I am glad I had the opportunity to see this work.

The effort put in by the authors is diminished by the way this work is presented. I list below a few aspects that justify my comment:

- The authors present in Chapter 4, Experiments, the outline of the study, in which you can also see the outline of the method used. This outline is more appropriate for Chapter 3, where the presentation of the research methods used is expected.

- In Chapter 3, the authors present the outline of the proposed method in Figure 3. This is unclear and difficult to understand. The use of Figure 4 may be more appropriate for what the authors intend to present in this chapter.

- In Chapter 2, entitled Related work, the authors present an analysis of some of the techniques on which their proposed method is based. The expectation created is to find out if there are related works that solve the problem the authors are considering. The analysis performed is useful, but not the most appropriate for this chapter.

- In Chapter 4, the authors present a set of results in tables and figures. These results are not clearly enough related to the methods used, the formulas that were used.

I suggest the authors take my comments very seriously and try to improve the manuscript to greatly strengthen the way they present their research.

Author Response

Dear Reviewer,

Thank for your kindness working and valuable advice.

We revised the manuscript carefully according to your comments and suggestions, and provide the point-by-point responses in submit word file. All changes were marked in red using the “Track Changes” function in the revised manuscript.

Best regards,

Xizhen Zhang

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report (New Reviewer)

Comments and Suggestions for Authors

The improvements made by the authors are consistent with my observations.

Author Response

Dear Reviewer,

we thank you very much for your kind work and valuable comments

We revised the manuscript carefully again, mainly focusing on the introduction, discussion,  conclusion, and reference. All changes were marked in red using the “Track Changes” function in the revised manuscript.

Best regards,

Xizhen Zhang

 

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

Comments and Suggestions for Authors

This is a well matured article, reporting a study that appears well designed and executed. There are small issues identified in the text in the document attached. The English needs improvements here and there, but overall the text is fluent.

I have two main concerns that prevent me from suggesting outright approval at this stage:

1. Relevant of results. Some of the results report improve on the state of the art only slightly. It is important to discuss their relevance and put them into context.

2. Reproducibility. There are no references to either data or source code. As it stands the study is effectively not reproducible

 

Comments for author File: Comments.pdf

Comments on the Quality of English Language

Some improvement needed.

Author Response

See attached

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

In this work, the authors have presented an adaptive deep multi-order spatial-spectral residual feature extraction method to mine the weak information from remote-sensing images. The above paper is logically written and contains the necessary background introduction, method analysis, and experimental verification. However, some specific problems should be improved to provide more convincing and clearer results for readers. Some specific comments can be seen in the attached file. 

Comments for author File: Comments.pdf

Author Response

See attached

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

In order to mine the weak information of remote sensing images, this manuscript proposes the deep multi-order spatial-spectral residual feature extractor, which considers global information and three-dimensional cube structures by combining low-rank representation, high-order residual quantization, and multi-granularity spectral segmentation theories. However, some concerns can be addressed before further review process.

1.    The description of the novelty is not enough.

1） Figure 3 is unclear, which impacts readers' reading. The figure and their corresponding descriptions are important for the paper's novelty to be understood, while the figures and their corresponding in this paper are unclear.

2) This paper is about weak information extraction, however, the description of weak information is lacking, and we can not capture the motivation of this paper.

3) The overall writing logic and expression of the paper need to be improved.

2. There are some problems in the experiments.

1) The experiments lack some compared methods. It can not prove the superiority of this algorithm.

2) The ablation experiments are not enough.

3. Some editing errors need to be improved.

1) The format of full-text application references should be consistent, e.g. ‘Kavitha et al. [37]

‘ and ‘Pang [41]’.

2) The DMSR and DMSC are not defined in the section of the abstract.

Comments on the Quality of English Language

I can understand his English writing.

Author Response

See attached

Author Response File: Author Response.pdf