MAAFEU-Net: A Novel Land Use Classification Model Based on Mixed Attention Module and Adjustable Feature Enhancement Layer in Remote Sensing Images

Round 1

Reviewer 1 Report

This study offers a classification model based on Mixed Attention module and Adjustable Feature Enhancement layer, and builds a dataset to evaluate the model.

 1. The proposed model structure is very similar to U-Net, and borrows the swin transformer and the attention mechanism.

 2. In terms of model construction, please specify the reasons for introducing the swin transformer and the attention mechanism. Why the swin transformer and the attention module are combined in parallel?

 3. Compared with the listed methods, the proposed method has a large improvement in MIoU. However, compared with other methods, the segmentation results of the proposed method do not seem to be better. It is better to present more segmentation results.

 4. There are many existing semantic segmentation networks. The listed methods in Table 4 are relatively old. It is better to present new methods.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Extensive literature review done. Well explained study objectives. Information on how the gwadar dataset was generated could be mentioned.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

This paper highlights the importance of accurate land use classification and introduces the MAAFEU-net model as a novel approach to achieving this. The Mixed Attention (MA) module and Adjustable Feature Enhancement layer are explained as the key components of the model that allow it to capture class-related features and enhance classification ability for similar types. However, the part of Adjustable Feature Enhancement layer is not well explained. Further information should be provided.

More revision suggestions are listed below:

Line 86: It appears that sentences are missing.

Line 126: This is the first time that “CBAM” is mentioned, so the full term should be indicated.

Line 170: More detailed information should be provided to explain how to utilize the Focus structure and why the Focus structure is particularly effective at mitigating this issue.

Line 241: for this section, it is unclear why the feature enhancement layer should be generated in this way. To make this section more convincing, more detailed information should be provided to explain X1 and X2 parameters. It is necessary to explain why it can fuse and strengthen the segmented type features. What does Conv2D represent in figure 4?? Every word used in a figure must be accompanied by a description. Major revision is needed for this section.

Line 259: how to remove unnecessary bare land and sea and why? Explain in more detail.

Line 377-386: in this paragraph, MAAFEU-net is mentioned. However, in Figure 6, the authors use SCCU-net instead of MAAFEU-net, which is very confusing. The description must correspond with the figures. The word “SCCU-net” is not mentioned in the whole manuscript.

Figure 7 also has this issue.

How does the Ground truth image generate? Is it the same with Labeled Image? The definition of ground truth is ambiguous and misleading, particularly when the words in the figure and the description do not match.

Line 467:  provide further insight into the findings, their implications, possible drawbacks, and future research directions.

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

Thank you for the opportunity to review the manuscript (MAAFEU-net: A novel land use classification model based on Mixed Attention module and Adjustable Feature Enhancement layer in remote sensing images) on this interesting and timely topic. This study focuses on a new method for land use information classification. The authors proposed a new land use classification model based on hybrid attention modules and adjustable feature enhancement layers, called MAAFEU net. It allows the model to focus on discriminative features of specific targets and capture features related to different land use types. In addition, an adjustable feature enhancement layer was proposed to further enhance the classification ability of similar types. The article fits in with the scope of this topic and has a complete structure. Of course, this work also has some places to be improved and next I would highlight those aspects so as to make the paper better. My comments are as follows, please consider and revise.

1. Abstract. Abstract needs to be simple but can also show the key points of the MAAFEU net. It cannot contain too much content, nor is it a repetition of the results. It needs to highlight the key content of the research (including the refinement of the results), which can better arouse the interest of readers.

2. This article may have been completed in a hurry, and may not have been carefully combed before submission. Many sentences need to be combed. For example, lines 145-150 should not be capitalized(“In this paper, two different attention mechanism modules, Swin Transformer [30] and CBAM [31], are used to build the MA module, which makes full use of the different advantages of the two modules for feature extraction and strengthens the connection between global and local features. As shown in Figure 1, the MAAFEU-net consists of three main components: the backbone network with the addition of Focus, the MA module to enhance feature extraction and the feature enhancement layer to further process the upsampled feature maps.”). The period symbol in line 29 is in Chinese format.

3. Introduction. The content of the introduction is not sufficient and needs to be reconstructed or supplemented. Of course, the author introduces the background of land utilization information and machine learning algorithms of land use classification, which is necessary. Please consider referring to and quoting the following documents to enrich the parts: https://doi.org/10.3390/land11010014, https://doi.org/10.3390/rs15071813

4. In the last part of the introduction, the author tries to introduce the innovation of this article, but the description is not clear. It is suggested to enhance the research significance of this article.

5. There are many abbreviations in the entire text, but they are not standardized. Please refer to the journal for revisions. For example, CNN, SVM.

6. The quality of the picture is very good. The conclusion are good.

7. The article needs to discuss the limitations and critical explanations of the MAAFEU net.

I believe that the article will be greatly improved after being modified according to these comments. Good luck!

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

There are also the following issues:

1. In the abstract, "And six state-of-art land use classification models are used for comparison." Six deep networks are used for comparison: MANet, DC-Swin, VGG16-based U-net, Deeplabv3+, SegNnet, and PSPNet. Among the existing methods of land use classification, these listed methods are not all the best.

2. The English writing of the article needs further examination and improvement.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

The author made careful revisions according to the reviewer's comments, and the quality has been significantly improved, basically meeting the publication requirements. However, in order to further improve the quality of the article and promote the reputation of the journal, the author should consider the following comments.

1. The article has relatively few references, with less than 40 references. As an innovative method article, this is obviously not enough. The author should read, supplement, and cite more extensive relevant literature as a basic support. Therefore, I strongly recommend the author to further supplement the references, and the author can consider the following references.

https://doi.org/10.1016/j.uclim.2022.101347 ;

https://doi.org/10.1016/j.scs.2020.102548 ;

https://doi.org/10.1016/j.jenvman.2021.112498 ；

https://doi.org/10.1016/j.jenvman.2007.06.028 ；

https://doi.org/10.1016/j.scs.2021.103296 ；

https://doi.org/10.1016/j.scs.2021.102760 ；

https://doi.org/10.1016/j.scs.2021.103479

2. The conclusion section can be further optimized and refined to display key content.

Author Response

Please see the attachment.

Author Response File: Author Response.docx