review-paper


Reviews:

===============================================

 It records all my papers reviews and certainly can thereby help me to improve my next 
 academic writing according those negative comments

MUMLP

Comments to the Author
This paper proposes a Multi-Scale U-shape Multi-Layer Perceptron (MUMLP) model. This method includes the designed MSC (Multi-Scale Channel) block and the UMLP (U-shape Multi-Layer Perceptron) structure. Some experiments are conducted to analyze the effectiveness of the proposed method. The main comments are

Reviewer: 1

  1. For the background, it lacks some key literatures about the hyperspectral image, which can refer to 10.1109/TGRS.2016.2583219 and 10.1109/LGRS.2019.2936652.
    
    1. Please rewrite the contributions to highlight the innovation. 
      
      1. Table I and II can be combined to show the classification results of each class. 
        
        1. For the experimental results, please enhance analysis the reason why the proposed method can achieve better results.
          
          1. Please demonstrate the effectiveness of each module in the proposed architecture.
            

Reviewer: 2

  1. Since all the experiments are about image classification, and classification has different meaning compared with identification, change your title.
  2. Main contribution of a paper should focus on the highlighted novelty, not the experimental results in numbers, refine the first two contributions and delete the last one.
  3. Move Fig.2 and Table 1 into next pages.
  4. Acknowledgement part is missing, at least acknowledge the data provider.

CUGCN

Associate Editor
Comments to the Author:
The expert reviewers have raised some major issues, particularly regarding the experimental section. Additionally, they asked for more clarification on the novelty of the proposed method compared with the state-of-the-art. I have some concerns myself that I mentioned below. I encourage the authors to apply the major changes and provide a point-by-point response letter addressing all the concerns.

Editor

  1. The datasets used are all old. Please evaluate your method using newer benchmark datasets such as Houston 2013 and 2018.
  2. Please provide the classification maps for all datasets. To save up space the authors can only select a few methods which provide the highest accuracies for the visual comparisons of the final maps.
  3. Please provide the hyperparameter analysis. How sensitive is the proposed method to the selection of hyperparameters?
  4. Please provide citations for the competing methods used in the experiments. Additionally, I recommend the authors provide a comparison with an advanced shallow feature extraction technique such as OTVCA (see “Feature Extraction for Hyperspectral Imagery: The Evolution From Shallow to Deep: Overview and Toolbox”).
  5. Please provide more detail regarding the selection of training samples.
  6. Please provide the processing time and compare it to the competing methods.

reviewer 1

This paper presents a new model, named the Context-Aware Attentional Graph U-Net (CAGU), to improve interclass variabilities and interclass similarities when classifying hyperspectral data. In general, the paper is interesting since it provides some technical novelties within the remote sensing field. However, there are still some points about the work that should be improved before publication.

  1. (Page 1, Line 25, Right) Please, define the acronyms before their first usage.

  2. (Page 2, Line 57, Left) The authors should better motivate the use of a U-Net model with respect to other classification architectures.

  3. (Page 2, Line 26, Right) Please, try to be more specific when detailing the contributions of the work. For instance, it is not a good idea describing Fig.1 or talking about the conducted experiments as contributions.

  4. (Page 3, Line 37, Right) Please, try to join the paragraphs of this section when they are very short.

  5. (Page 3, Line 53, Right) Do the authors think that it would be possible to consider alternative kernel sizes? Please, provide more information about this.

  6. (Page 4, Line 20, Right) Please, try to better clarify the GRU model here. I think the writing is a little obfuscated. It would be also good to include some visual data here.

  7. (Page 5, Line 25, Left) The considered sampling strategy is not clear. What is the rationale behind it?

  8. (Page 5, Line 44, Left) Please, provide the references of the considered competitors.

  9. (Page 5, Line 52, Right) Please, explain here the insights why the proposed approach is outperforming the other competitors.

  10. Please, revise the whole writing carefully:
    (Page 3, Line 54, Right) “Copy and crop operation combines”
    (Page 4, Line 30, Left) “an Tesla”

Reviewer 2:

Comments to the Author

According to the submitted manuscript, my concerns are as follows.

  1. The full classification maps are expected to be given.
  2. Why is there only a classification map of Salinas dataset? If possible, please give the classification maps of other datasets.
  3. In Salinas dataset, why is the 2nd order CAGU higher than the 1st one with 10% of the sampled data? There is no such situation in the Pavia data set and the Indian pines data set?
  4. Please give the accuracy and loss function curves of the training and validation sets of the Pavia data set and the Indian pines data set.
  5. In the experiment, the latest hyperspectral image algorithm based on graph convolutional network should be added to show the superiority of this algorithm.

Author: Moule Lin
Reprint policy: All articles in this blog are used except for special statements CC BY 4.0 reprint polocy. If reproduced, please indicate source Moule Lin !
  TOC