DSSM: A Deep Neural Network with Spectrum Separable Module for Multi-Spectral Remote Sensing Image Segmentation
, Letong Han 1, Hongfei Fan 1, Zeju Wang 1, Bowen Du 1,2,*, Sicong Liu 3 and Qin Liu 1Round 1
Reviewer 1 Report
In this work a network structure is proposed and deeply detailed to perform feature extraction in MS images using DSSM to improve the representation of spectral features.
The experimentation cannot be considered as a main contribution since it is used to demonstrate the effectiveness of the proposed contributions.
In line 180, the text should indicate where the shortcoming of CNNs are illustrated with a text reference.
Figure 1 caption should be more descriptive. Also, some text size should be incremented.
The paragraph starting in line 215 should be rewritted. The concepts are not clear and the text is not properly written.
During all text and specially in Section 3.1.4, some nomenclature abbreviations are repeated and not coherent. Also, the computational cost should be clearly defined by the number of operations or speedup.
I consider the experimentation clear and complete. Finally, it is neccesary to revise the english writting during all text.
Author Response
Dear reviewer, thank you so much for your significant suggestions of major revision. The detailed revision note and the revised-version manuscript are in the attachment, please check to see if there are other problems. Thanks again for reviewing！
Author Response File: 
Author Response.pdf
Reviewer 2 Report
In Table 3 (around line 442) - under GPU consumption - what is unit of measure is "M"? I think that should be explained to understand how your method is more efficient
-Also as more of a major concern - in Section 3 Methodology - is there an interest in making the code available with data for use to the general public/scientist to replicate?
Or is there a plan to on a code repository that can be accessed publicly? It is unclear on what specific computing tools were used to conduct this. I think this should be added in case anyone would want to replicate the findings of the research.
Author Response
Dear reviewer, thank you so much for your significant suggestions of major revision. The detailed revision note and the revised-version manuscript are in the attachment, please check to see if there are other problems. Thanks again for reviewing！
Author Response File: Author Response.pdf
Reviewer 3 Report
The manuscript presents a deep neural network for semantic segmentation (which is referred to as DSSM). The proposed approach is based on spectrum separable module (SSM), which allows to extract more informative spectral features. The presented technique has been evaluated using two well-known datasets. The results show that the segmentation accuracy of the fusion strategy in the SSM is higher than other processing strategies. However, this result is achieved at the expense of increasing by 20% the GPU consumption. At this point, I think that the authors should include this data in the abstract because they say there that "the proposed approach improves the segmentation accuracy without increasing GPU consumption".
In order to verify the overall performance of the DSSM, the authors have compared their approach to other methods. The segmentation accuracy of DSSM is almost always better than others.
The manuscript is well-written and explains clearly the work done by the authors. The proposed technique is compared with the state-of-the-art techniques, obtaining good results. I only have some doubts and suggestions:
- I miss details about the GPU implementation.
- Why is the reason not to use the same benchmark for the two kinds of comparison?
- How robust are the results using only one benchmark for each comparison?
- I would be interesting to include a comparison of the execution time between the methods.
- Some acronyms are not defined (eg., CVI or CV)
Author Response
Dear reviewer, thank you so much for your significant suggestions of major revision. The detailed revision note and the revised-version manuscript are in the attachment, please check to see if there are other problems. Thanks again for reviewing！
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The article is suitable for publication.
Reviewer 2 Report
Edits look good. Thank you. Only suggestion I would have is that the code comments you have under your python - if that can also be translated to English maybe more people can use it without needing to use a translator?
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.
