Scene Classification Based on Heterogeneous Features of Multi-Source Data
Round 1
Reviewer 1 Report
The manuscript presents a scene classification model that integrates heterogeneous features of multi-source data. Extensive experiments are conducted and many state-of-the-art methods are compared with the proposed method. I have some major comments:
1. In the Introduction, many references to multi-source remote sensing data are not cited. Such as the following literature [1] [2]. Please cite the two references [1] [2] to make this section completed.
[1]Y. Fang, P. Li, J. Zhang and P. Ren, "Cohesion Intensive Hash Code Book Coconstruction for Efficiently Localizing Sketch Depicted Scenes," in IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1-16, 2022, Art no. 5629016, doi: 10.1109/TGRS.2021.3132296.
[2]Y. Sun et al., "Multisensor Fusion and Explicit Semantic Preserving-Based Deep Hashing for Cross-Modal Remote Sensing Image Retrieval," in IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1-14, 2022, Art no. 5219614, doi: 10.1109/TGRS.2021.3136641.
2. Please explain why the proposed algorithm employs the Lie Group manifold space distance instead of Euclidean space distance. The advantage of the Lie Group manifold space distance should be added.
3. The introduction section should distinctly introduce the “Multi-source”, such as which source the data is obtained from.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The authors address the problem of remote sensing scene qualification in the application of deep learning programs. They propose a new scene classification model that integrates heterogeneous properties of multi-source data. The approach of the paper is easy to follow for a reader, although the clarity of the English text leaves much to be desired. The authors demonstrate by means of various tests that their modeling approach has advantages over the previous approach. The paper has a sufficient number of citations, a sufficient number of explanatory tables and a very good quality of its figures.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
This paper presents a novel scene classification model that integrates heterogeneous features of multi-source data. Specifically, a multi-granularity feature learning module and the maxout-based module were proposed for the framework. And the Lie Group Fisher algorithm was used for scene classification. The proposed framework was analyzed explicitly with a range of examples. Although generally favorable for publication, I suggest considering the following comments.
1. In the introduction, the author summarized mainly three models about the scene classification task, but the writing is too abbreviate. More introduction to scene classification in remote sensing (e.g., scene classification for high spatial remote sensing image) would be helpful.
2. In the introduction, the author stated “R. N. Marandi and H. Ghassemian[19] proposed a joint feature representation model. However, the above feature fusion methods do not fully consider the correlation and redundancy of features.” It is suggested that the authors introduce more joint feature representation models and explain the advantages and disadvantages to introduce the innovation of this paper.
3. The author presented three challenges, it is recommended that the content represented by challenges 1 and 2 be shown in images.
4. On the second page of the article, the author mentioned “ontology structure”, please introduce “ontology structure”.
5. The literature review about deep learning-based RSI scene classification and integrating knowledge for RSI scene classification in recent 3 years should be added, such as doi.org/10.3390/rs14061478ï¼› doi.org/10.1016/j.rse.2022.112916.
6. In Figure 5., the author described "The principle of dense module.", but the image does not clearly show the structure of the dense module. It is recommended to clearly label the links and layers in the image.
7. In the results, please explain the reasons for choosing these advanced networks as comparison experiments.
8. Suggest adding confusion matrix of other advanced methods for better analysis of experimental results.
Comments for author File: 
Comments.docx
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Well revised. Acceptance recommended.
