Dynamic Convolution Self-Attention Network for Land-Cover Classification in VHR Remote-Sensing Images
, Yue Zhang 2,3, Tao Lei 2,3,*, Yingbo Wang 2,3, Yujie Zhai 2,3 and Asoke K. Nandi 4,5Round 1
Reviewer 1 Report
Comments for author File: 
Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
In the paper, a method for land cover classification in very high-resolution (VHR) remote sensing images is presented addressing information redundance and resource usage reduction. Comments:
1. Please extend the VHR abbreviation in the abstract and introduction.
2. Surprisingly, the approaches [48-53] compared in the experimental section are not introduced in Related Works. Do they belong to the state-of-the-art (SOTA)?
3. The “negative effect of useless feature maps” is mentioned only in the abstract and the list of contributions. The redundant maps should be shown and their effect studied in relation to the performance of the approach. The paper lacks such experimentation.
4. The importance of the size of the receptive field should be investigated in detail.
5. Some recent works that outperform the presented method (e.g., Buildings in Potsdam dataset in “Ferrari, Luca, et al. "Integrating EfficientNet into an HAFNet Structure for Building Mapping in High-Resolution Optical Earth Observation Data." Remote Sensing 13.21 (2021): 4361.”) are not cited. Some of them clearly outperform the introduced solution (“Nie, Jie, et al. "MIGN: Multiscale Image Generation Network for Remote Sensing Image Semantic Segmentation." IEEE Transactions on Multimedia (2022).”). Please see that the MIGN approach with its OA and F1 of 0.9326 and 9401 on the Vaihingen dataset, respectively. They are better than the reported 0. 9058 and 0.8881. Also, as reported in the MIGN paper, many approaches exceed F1 of 0.9058. Hence, the proposed method is not compared with SOTA.
6. The paper should also report the ioU metric, as it is often done in similar works.
7. The paper requires proofreading (e.g., page 8, line 330 “We”).
8. Please add the year of introduction in brackets next to each SOTA approach in lines 437-439 (page 11). Methods from Related Works are not considered in experimental evaluation.
9. The paper should contain a link to a webpage with source code ensuring the repeatability of the results.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Please find attached.
Comments for author File: Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
Please see the attachment.
Comments for author File: Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The previous comments have been considered. I have no more suggestions.
Reviewer 2 Report
The revision is satisfactory. Surprisingly, authors were able to address all issues I identified. The most convincing is the addition of the link leading to the source code of the method.
Reviewer 4 Report
The authors have answered my all concerns. I have no other question.
