Deep Learning for Remote Sensing Image Scene Classification: A Review and Meta-Analysis

Round 1

Reviewer 1 Report

Deep Learning for Remote Sensing Image Scene Classification: A Review and Meta-Analysis

·        In the abstract please highlight the importance of this study, and why it was conducted. A brief discussion about the benchmark datasets highlighted in the study, what are the significance of this study, open challenges, and future research scope. This will attract the readers to further read the paper in detail.

·        At the end of the introduction section please list the major significant contributions extracted from this study.

·        Figure 3 should resemble remote sensing work. Please redraw it.

·        Section 2.2 can have many methods based on CNN. Please explore more and show the different approaches handled by the authors. Show it pictorially and technical details in tabular format.

·        Section 2.2.1 Each subsection should be supported by tables with technical details and figures to represent equations if possible.

·        Section 2.3 should be supported by figures for Gan and Equations.

·        Section 3.4 needs more technical details. What techniques were applied for data augmentation? Most popular and effective techniques etc.

·        Table 4 can be categorized based on the approaches. Each approach should be separately displayed with discussion.

·        A separate section with the name open challenges is needed. List all the major challenges in this field of results.

No Comments

Author Response

Please see the attachment for the response to you comments. Thank you!

Author Response File: Author Response.pdf

Reviewer 2 Report

Deep Learning for Remote Sensing Image Scene Classification: A Review and Meta-Analysis:

·        Add some of the most important quantitative results to the Abstract.

·        These sentences are not necessary and should be deleted: “The remainder of the paper is organized as follows: Section 2 provides a detailed review of remote sensing scene and feature extraction processes; Section 3 illustrates the meta-analysis on related papers; Section 4 discusses the findings and future directions; and Section 5 presents the concluding remarks.”.

·        In the last paragraph of the Introduction, the authors should mention the weak point of former works (identification of the gaps) and describe the novelties of the current investigation to justify that the paper deserves to be published in this journal.

·        Why didn’t you use other databases such as Web of Science and/or Google Scholar? Why Scopus?

·        Why isn’t America involved in Figure 5?!

·        “Furthermore, the meta-analysis reveals that CNN-based approaches consistently outperform GAN-based approaches in terms of classification accuracy due to their supervised nature.”. Explain.

At the end of the manuscript, explain the implications and future works considering the outputs of the current study.

Author Response

Please see the attachment for the response to your comments. Thank you so much!

Author Response File: Author Response.pdf

Reviewer 3 Report

The paper presents an extended up-dated review on remote sensing scene classification using deep learning methods, along with a novel meta-analysis on this topic. The proposed approach and the obtained results are interesting, useful, reliable and valuable, using an excellent English style and logical approach, but the paper should be carefully reviewed for the sake of clarity.

The following issues are recommended to improve the paper:

1.     Define all acronyms at their first use in the Abstract and body text, respectively, even they are well-known in literature. E.g., CNN, GAN, etc. Check carefully the entire manuscript for other similar cases of undefined acronyms.

2.     Recommendation to include a Nomenclature section for the used acronyms.

3.     Fix several typing mistakes, e.g., “(i) All 45 classes” and “(ii) The 45 sub-levels” – use “(i) all …” and “(ii) the…”, see other similar cases p. 9; “mechanism: In the”, use comma “,” between items in an enumeration (before “(ii)”), etc.

4.     “All architectures are evaluated using the overall accuracy (OA) metric to facilitate a comprehensive comparison.” – detail shortly the specified metric OA.

5.     Conclusions: please address also the novel approaches proposed/involved since 2020 compared to previous period; future research directions may be also highlighted here.

Author Response

Please see the attachment for response to your comments. Thank you so much!

Author Response File: Author Response.pdf

Reviewer 4 Report

Well done!  You have done an excellent job of systematically summarizing and concluding the new progress in the scene classification of remote sensing images. The whole work is well-written and organized.  In my opinion, this paper can promote the development of this field. 

A suggestion is that this paper is wordy and long pages. Authors can use more concept figures or diagrams to well illustrate the progress and research interests.

Author Response

Please see the attachment for the response to your comments. Thank you so much!

Author Response File: Author Response.pdf

Reviewer 5 Report

This review categorizes remote sensing image scene classification into two major categories: CNN-based and GNN-based methods. However, Visual Transformers (ViT), as a pure Transformer architecture, can perform image feature extraction without relying on convolutional layers. Scene classification based on ViT has also received considerable attention from researchers, yet this review overlooks this aspect. Therefore, I suggest reconsidering the publication of this review until the authors include relevant content about ViT. Below are some recommended references for your consideration.

[1] Bazi Y, Bashmal L, Rahhal MM, Dayil RA, Ajlan NA. Vision transformers for remote sensing image classification. Remote Sensing. 2021 Feb 1;13(3):516.

[2] Xu K, Deng P, Huang H. Vision transformer: An excellent teacher for guiding small networks in remote sensing image scene classification. IEEE Transactions on Geoscience and Remote Sensing. 2022 Feb 17;60:1-5.

[3] Lv P, Wu W, Zhong Y, Du F, Zhang L. SCViT: A spatial-channel feature preserving vision transformer for remote sensing image scene classification. IEEE Transactions on Geoscience and Remote Sensing. 2022 Mar 8;60:1-2.

[4] Zhang J, Zhao H, Li J. TRS: Transformers for remote sensing scene classification. Remote Sensing. 2021 Oct 16;13(20):4143.

[5] Tang X, Li M, Ma J, Zhang X, Liu F, Jiao L. EMTCAL: Efficient multiscale transformer and cross-level attention learning for remote sensing scene classification. IEEE Transactions on Geoscience and Remote Sensing. 2022 Jul 28;60:1-5.

Author Response

Please see the attachment for the response to your comments. Thank you so much!

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have revised the paper according to the reviewers comments and hence can be accepted for publication.

No COmments

Author Response

Thank you for your positive comment.

Reviewer 2 Report

I appreciate the authors addressing the comments. The manuscript can be accepted in its current form. Congrats!

Author Response

Thank you for your positive comment and accepting the manuscript. 

Reviewer 5 Report

I appreciate the authors for the significant improvements of the manuscript. However, I still suggest listing the pure visual transformer as a primary method for remote sensing scene classification (lines 5-6) rather than as one of the CNN-based methods (lines 255-293), even though there are some hybrid methods of transformers and CNNs. Additionally, a minor issue is in Figure 3, the spacing between each rectangle should be consistent to make it look more aesthetically pleasing.

Author Response

Please see the attachment for the response to your comments. Thank you so much!

Author Response File: Author Response.pdf