Residual Attention Mechanism for Remote Sensing Target Hiding

Round 1

Reviewer 1 Report

In this study, the authors propose a Residual Attention Target Hiding (RATH) model for remote sensing target hiding based on deep learning. While the authors did the hard work still have some queries and suggestions to improve the manuscript:

 

1.       There seem to be some non-standard writing and grammar or syntax error

(1) Lines 41-46, when some terms first appear, such as " Cont Atten ", "Gated Conv" and "Partial Conv", they should be in the form of "full name (abbreviation)" rather than directly using abbreviations.

(2) Line 43, please replace "the missing area of the image" with "the missing areas of the images".

(3) Line 239, please replace "the Fuse module" with "the fuse module".

(4) When describing the inpainting operation on an image, the authors should use the term "inpainting", but the word "repair" is used in multiple places in the paper, which is an incorrect term.

(5) Line 151, in sentence “Compared to Partial Conv [9] and Gated Conv [8], gated convolution offers the following advantages: …”, “Gated Conv” and “gated convolution” should refer to the same attribution. The “and Gated Conv [8]” should be deleted.

Some similar problems should be checked throughout the manuscript.

2.       Lines 54-56, residual connections can indeed alleviate gradient explosion and dispersion, which can promote the convergence of deep networks. However, this seems to have nothing to do with solving overfitting problems, which are generally alleviated through techniques such as dropout, regularization terms loss and data augmentation.

If the authors believes that the proposed model can “be trained extensively without overfitting”, it would be better to provide evaluation curves for both the training and validation sets.

3.       Line 71, there seems to be an inconsistency between the description “on diverse datasets” and the Section 4, as the authors introduces that the dataset used in this paper was produced using the Mnih Massachusetts Building Dataset in the Section 4.

4.       The related work is all about Inpainting, and does not involve the research work of remote sensing and target hiding.

5.       In section 3.1, the introduction to the overall structure is somewhat confusing and lengthy. It is best to provide a more concise and clear description of the main framework, and then gradually explain the design concept.

6.       Section 3.3 seems to have been moved to Section 4 as a description of the experimental setup.

7.       It is recommended that this paper provide the results of ablation experiments to verify the effectiveness of key module.

8.       There are some issues with figures, tables, formulas, and their descriptions.

(1)    In Figure 1, there is no legend or explanation for the red rectangle in the dashed box at the bottom left corner, and its color is not distinguishable from other modules.

(2)    The symbols of formula (2) are not fully explained, and the symbols of the two activation functions are opposite to that shown in Figure 2 (b).

(3)    Section 3.2.2 introduces too much content on Contextual Attention layer, but it seems that this is not the original work of this paper. Please simplify this section and make the improvement of CAL the main content. Meanwhile, please present the improved results of CAL in Figure 3 to highlight the work of the paper.

(4)    For Figure 4, please replace "Ground Truth" with "Ground Truths"

(5)    For Table 1, the title of the table is “two image inpainting methods”, but there are three methods involved in the Table 1. Then, the abbreviations of the proposed method are inconsistent in Tables 1 and 2. Finally, as shown in Table 1, the complexity and training speed of “Self-Atten” and “Res Atten” are the same, and it seems necessary to further introduce the advantages of “Res Atten” over “Self-Atten”.

(6)    Line 323, The description “Our proposed method exhibits the smallest values on ℓ1 Sim and ℓ2 Sim indices” is not consistent with Table 2.

(7)    The results shown in Figure 6 and Figure 10 are inconsistent with the description in the figure title. order of schemes a and b is opposite.

(8)    The title of Figure 8 does not mention the comparison method. Please modify it according to Figure 7.

(9)    The title and content of Figure 9 are inconsistent.

(10) The authors point out “lack of accepted evaluation methods for target hiding” and believe that the lowest SSIM and UQI represent the best target hiding performance. However, the lower these indicators, the worse the inpainting effect. It would be better if the authors could provide a more comprehensive explanation of this, or provide relevant literatures as a basis.

(11) Line 387, please replace "red region" with "red box".

(12) Why did the proposed method not continue to be compared with other methods in Table 4, but only with Gated Conv.

There are a number of typos in the paper. The authors must improve the writing of the English.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The author of this manuscript, " RESIDUAL ATTENTION MECHANISM FOR REMOTE SENSING TARGET HIDING " presented some stunning experimental results. However, some comments are there to fix it before publication.

1.       It would be better to have a Graphical Abstract. A well-drawn graphical abstract promotes readers' understanding.

2.       The abstract should identify the research gap and propose the academic contribution of the paper, showing numerical results. The number of words should be controlled to about 200-400.

3.       The author's introduction needs to be optimized, and we suggest that the author evaluate what needs to be improved in the introduction according to the following criteria.

·       What is the problem to be solved?

·       Are there any existing solutions?

·       Which is the best?

·       What is the main limitation of the best and existing approaches?

·       What do you hope to change or propose to make it better?

·       How is the paper structured?

4.       When the authors cite references, most of them are simply displayed in the paper. I suggest that the authors relate these references to the work of this paper, for example, how certain studies have influenced their work.

5.       In the related work, the authors must provide the design principles and basic knowledge of technical solutions. Some new and important work that uses deep learning methods to optimize image recognition needs to be mentioned. These papers can provide some theoretical basis, and some suggestions are as follows: https://doi.org/10.3390/land12040831, https://doi.org/10.3389/feart.2023.1182145, https://doi.org/10.2352/J.ImagingSci.Technol.2023.67.3.030402, 10.1109/MGRS.2022.3145854, https://doi.org/10.1016/j.earscirev.2022.104110

6.       How the comparison manifests, it looks like figures 4 and 5 are the same.

7. Figure 9 does not seem to match what is described in the text.

It is suggested that the authors have someone competent in the English language and familiar with the subject of your manuscript go over the manuscript and polish it. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

1.    1. The  abstract must have precise quantification. It reads vaguely about efficiencies over prior works. The authors must put the benchmarks carefully and quantify the performance needs a major revision here.

2.     Each acronym should be explained the first time it appears in the text, even if it appeared in the abstract. Check all abbreviations in text: each word should start with capital to explain an abbreviation.

3.     Suggest  the authors to provide a comparative analysis of the various techniques as used in the literature survey, preferably in a table to justify the outcome as claimed to be very efficient, where there is a small confusion.

4.     Redraft all the titles and contents appropriately – Need Proper numbering and data flow.

5.     The experimental results can be more organized to validate the theoretical ideas – Need a clarity.

6.     Suggest the authors to follow a standard template to present their ideas in a much-organized manner - The flow of context of information is very poor/ fuzzy, the titles of the figures to be more precise.

7.     Redraft the  sections 5, 6  addressing the result findings appropriately, include discussion section. furthermore  improvise the conclusion section with at least one future scope  in specific and not generalized.

8.     It is required to have the latest citations at least to be included under literature survey to justify the results been compared to the current results.

9.      Why authors have selected the RESIDUAL ATTENTION MECHANISM FOR

REMOTE SENSING TARGET HIDING as most efficient. Need to  provide the justifications or reasons for the same.

10. The grammar and syntax of many sentences are incorrect containing numerous grammar mistakes, which should have been easily detected and corrected by more careful and thorough proof-reading

The grammar and syntax of many sentences are incorrect containing numerous grammar mistakes, which should have been easily detected and corrected by more careful and thorough proof-reading at all sections.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Most of the comments are addressed properly. Manuscript can be accepted in present version.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The Submission has been greatly improved and is worthy of publication.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

1.     All keywords (abbreviations) must be mentioned in the abstract which are still at large and missing.

2. The titles of the figures need to be precise and not like paragraphs.

3. Section 4: Need to revise again, while you have conducted several tests, the title seems to be vague- need clarity in this section

4. Contents from section 5 are required to be included in section 6- fuzzy by nature again

5.    Further,  as per the article contents there still exists revision in English grammar and word choice as used- use of we at multiple sections.

 

while re- reviewing the manuscript, it is still found that the authors need to reassess  the contents as there still exists revision in English grammar and word choice as used.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf