Hybrid Attention Based Residual Network for Pansharpening

Round 1

Reviewer 1 Report

General considerations:

• The proposal seriously and clearly addresses a specific contribution that can improve the processing of satellite imagery. The research itself is not a revolutionary proposal, but it is a respectable new step in the discipline.
• The presentation of the contents is very good, and the graphic work of the figures that establish the different analytical comparisons is outstanding. As a very minor issue, it is recommended that the font of the figures should coincide with that of the text of the manuscript. Likewise, in a document with a total length of 16 pages, the proportion of figures, tables, formulas and references seems very appropriate for reading and comprehension.
• The document contains a total of 33 employed references, of which 19 are publications produced in the last 5 years (58%), 9 in the last 5-10 years (27%) and 5 than 10 years old (15%), implying a total percentage of 85 % recent references. In this way, the total number of references used can be considered improvable, but their actuality is very high.
• Finally, the quality of the English is good, and most importantly, the structure of the ideas is very good. Despite the complexity of the subject matter, the texts are clearly and rigorously explained.

Title, Abstract and Keywords:

• The title is in line with other approaches commonly used in research related to this topic. In this case, it is difficult to make suggestions for improvement.
• The abstract is correct, however, it is recommended that the authors specify what the improvements achieved and the increase in effectiveness achieved by the research carried out consist of.
• It is recommended to replace the keywords with those that currently match the terms that appear in the title of the proposal. This broadens the thematic scope of the research.
• Finally, authors should try to incorporate into this block (title + abstract + keyword) the acronyms that are later key to identifying the concepts throughout the reading of the manuscript.

Section 1: Introduction

• This is one of the clearest and best-organised introductions I have read on this subject. I congratulate the authors on this.
• I would like to emphasise the clarity with which the ideas and technical concepts are explained, without leaving any questions unanswered.

Section 2: Related work

• As in the previous section, section 2 provides a detailed review of the main methods used for the topic of work, as well as their pros and cons. The text is good.

Section 3: Proposed method

• Overall, the methods and formulas used are presented in a coherent and rigorous manner.
• At a particular level, the mention included between lines 130 and 138 could be better integrated into the manuscript, as a table or even as a nomenclature section.

Section 4: Experiment and Discussion

• The comparison of the results obtained with the research proposal is very complete and demonstrates its real contribution to the field of study analysed. It is a section which, like the previous ones, has little scope for improvement.

Section 5: Conclusion

• The conclusions section considers future lines of research but does not provide reflections on the applicability of the achievements to real situations.

Final evaluation

In summary, the research is interesting and provides valuable results. Thus, I consider that the manuscript is of sufficient quality to be published in its current state.

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

The article content is aligned with Remote Sensing Journal. The authors proposed a deep learning-based method for pansharpening using an attention mechanism. This is a hot topic in remote sensing image analysis.

 

My recommendation is for a major revision based on the following comments and suggestions.

 

The paper has content to be considered an “article”. However, it has only 16 pages. For this type, it is required 18 pages. Please, see https://www.mdpi.com/journal/remotesensing/instructions. The notations, for example, can be expanded and included in a section at the end of the article. More qualitative analysis can be presented. And also, more recent references (2020-2021) can be added.

 

First, be careful with the following statement: “The existing pansharpening methods suffer from the problems of spectral distortion and lack of spatial detail information, which might prevent the accuracy computation for ground object identification”. All the current methods did not pay attention to this point? Maybe, you can use “in general”.

 

The abstract did not present details regarding the conducted experiments. I suggest including details regarding the data sets.

In the Keywords, it can be considered words different from the title.

In the first paragraph of the introduction, it was mentioned three types of satellites. As a sentence for a general contextualization, I suggest not include those names. There are other satellites that were not mentioned.

In Figure 1, which method was used? A legend is necessary to better understand. Please, explain the reason for this distortion.

The methodology and results are clear.

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

Manuscript title: Hybrid Attention based Residual Network for Pansharpening

The author submitted a Hybrid Attention based Residual Network for Pansharpening. The presented menu script proposed a hybrid attention based residual network for pansharpening. The idea of using a hybrid attention mechanism on the basis of a residual network is interesting but there are some major issues in the current version of the manuscript. So, I recommend major revision decision by considering the following comments and suggestions.

1. In the abstract section, some of the long sentences should be revised to avoid redundancy and extra explanation.
2. Keywords in the manuscript are not in alphabetical orders. Please write it alphabetically.
3. All abbreviations should be defined well. Define them at the first appearance and use the short form accordingly. (e.g. Deep Residual Attention Convolutional Neural Network “DRANN”)
4. In Experimental Results, the time complexity and FPS of the proposed model are also not discussed. The authors should discuss it for the proposed model comparatively and briefly.
5. In Experimental Results, please provide training, validation accuracy and loss graphs of the proposed model as well as the confusion matrix of each dataset.
6. In Experimental Results, only one “CUHK-PEDES” dataset is used for the experiment. It would be great full to see experimental results on a minimum of 2 datasets. Kindly perform your experiments on one another dataset as well.
7. It is recommended to include the most recent research articles and their references. "An Efficient Anomaly Recognition Framework Using an Attention Residual LSTM in Surveillance Videos" Try to include mainly peer-reviewed reputed journal and top-tier conference in this area. The format of references is also not uniform.

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 3 Report

The authors have addressed my comments well I recommend it for publication now.