DHCAE: Deep Hybrid Convolutional Autoencoder Approach for Robust Supervised Hyperspectral Unmixing

Round 1

Reviewer 1 Report

In this work, the authors proposed a method for robust supervised Hyperspectral unmixing based on a deep hybrid convolutional autoencoder network. This is an interesting study with promising results.

The results and conclusions are reported clearly and concisely.

However, the authors should make some clarifications regarding the application of the method used. In more detail, authors should report the specifications of the adopted computer and the description of the used development environment and / or the adopted software in the methodology section.

These information regarding the implementation of the proposed approach are very important to allow other scholars to replicate the study.

English language and style are fine. However, the text would need some minor corrections and editing.

Detailed comments are provided in the attached file.

Comments for author File: Comments.pdf

Author Response

Dear Reviewer,

As per the comments, necessary changes have been incorporated into the revised version. The authors owe gratitude to the anonymous referee for his/her valuable comments and suggestions, which have helped us improve this paper. We have tried our best to address all concerns and comments. Our response to your comments is attached for your kind consideration.

 

Thanks

 Yours Sincerely 

Fazal Hadi, Jingxiang Yang, Matee Ullah, Irfan Ahmad, Ghulam Farooque, Liang Xiao*

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper presents an AE methodology utilizing 3D and 2D convolutions to solve the hyperspectral unmixing problem. The proposed methodology is compared with other methodologies on one synthetic and on three real-world datasets. The paper is mostly well-written, although it still needs to be proof-read. In general, all the captions need revisiting to be standalone.

Major remarks:

* No line numbering; it's difficult for us reviewers to point you to the proper place!

* The introduction of the other approaches needs to be slightly improved by expanding very briefly on what each of the other methodologies does (e.g. second paragraph of page 2).

* I am missing a discussion on how you selected this architecture. Why not use more (fewer) layers? Why 3 3-D and 2 2-D layers? Why is the decoder part so "simple" compared to the encoder? Usually there is a symmetry in autoencoders.

* please tell us the total number of parameters of your network

* the novelty needs to come forward: is it the combination of 3-D and 2-D simultaneously?  Please be more explicit in the introduction

* Discussion for future work is missing

* A more critical view of the methodology is missing. What are the drawbacks?

 

Minor remarks

* Figure 3 please provide a legend to tell us what each of these end-member signatures is

* DHCA should be defined on the first in-text mention (besides abstract)

* In 2.1, shouldn't X be 3-D (two spatial one spectral)?

* Figure 2; the reconstructed image appears to be smaller than the original, is this intentional? If the dimensions are the same then you should consider placing the same image on the right (in terms of size).

* gorund => ground

* Figure 5 and elsewhere what is DHCAE? Either DHCA or DHCAE, right? Am I missing something?

 

Author Response

Dear Reviewer,
As per the comments, necessary changes have been incorporated into the revised version. The authors owe gratitude to the anonymous referee for his/her valuable comments and suggestions, which have helped us improve this paper. We have tried our best to address all concerns and comments. Our response to your comments is attached for your kind consideration.
 
Thanks
 Yours Sincerely 
Fazal Hadi, Jingxiang Yang, Matee Ullah, Irfan Ahmad, Ghulam Farooque, Liang Xiao*

Author Response File: Author Response.pdf

Reviewer 3 Report

See attached file.

Comments for author File: Comments.pdf

Author Response

Dear Reviewer,
As per the comments, necessary changes have been incorporated into the revised version. The authors owe gratitude to the anonymous referee for his/her valuable comments and suggestions, which have helped us improve this paper. We have tried our best to address all concerns and comments. Our response to your comments is attached for your kind consideration.
 
Thanks
 Yours Sincerely 
Fazal Hadi, Jingxiang Yang, Matee Ullah, Irfan Ahmad, Ghulam Farooque, Liang Xiao*

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The authors have done a commendable effort to improve the manuscript. Still, some points need to be addressed in my opinion:

* The legend of figure 3: please include which materials these are. This is what I meant in my original review.

* Caption of Table 4 needs to be corrected grammatically "on the urban dataset remove noisy bands"

* Some of your replies to my remarks need to be incorporated into the manuscript, e.g. how you optimized the model structural and learning hyperparameters to derive in your architecture, what the total number of trainable parameters in your final network is.

* Since section 3 presents only the results, and section 4 lays out the conclusions, in my opinion you should either include a separate "Discussion" section or inlcude it in section 4 "Discussion and conclusions". Some points to discuss are the drawbacks of the purposed methodology, what could have been done differently, the suggestions for future work you included in Section 4 and the ones provided by reviewer 3, etc. A short but in-depth discussion of what the paper results are is crucial.

 

 

Author Response

Dear Reviewer,
As per the comments, necessary changes have been incorporated into the revised version. The authors owe gratitude to the anonymous referee for his/her valuable comments and suggestions, which have helped us improve this paper. We have tried our best to address all concerns and comments. Our response to your comments is attached for your kind consideration.
 
Thanks
Yours Sincerely 
Fazal Hadi, Jingxiang Yang, Matee Ullah, Irfan Ahmad, Ghulam Farooque, Liang Xiao*

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors answered all my questions.

Author Response

Dear Reviewer,
As per the comments, necessary changes have been incorporated into the revised version. The authors owe gratitude to the anonymous referee for his/her valuable comments and suggestions, which have helped us improve this paper. We have tried our best to address all concerns and comments. Our response to your comments is attached for your kind consideration.
 
Thanks
Yours Sincerely 
Fazal Hadi, Jingxiang Yang, Matee Ullah, Irfan Ahmad, Ghulam Farooque, Liang Xiao*

Author Response File: Author Response.pdf