LARS: Remote Sensing Small Object Detection Network Based on Adaptive Channel Attention and Large Kernel Adaptation

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The paper addresses an important and challenging task in object detection, specifically focusing on small object detection in remote sensing images. The proposed method combines adaptive channel attention and large kernel adaptation to improve localization accuracy, which is a significant contribution to the field.

1.The paper would benefit from improved clarity in several sections, particularly in describing the adaptive channel attention and large kernel adaptation blocks. Clearer explanations of how these blocks operate and interact within the network architecture would enhance understanding for readers.

2.The paper introduces a layer batch normalization method to address recognition confusion. Further insights into how this method improves the model’s performance, especially in the context of small object detection, should be elaborated. Clear evidence or analysis showing its effectiveness in reducing misclassifications would strengthen this aspect of the paper.

3. While the experimental results show state-of-the-art performance, it would be beneficial to include a more comprehensive comparison with other recent methods in small object detection in related work section. 

[1] Single-frame Infrared Small Target Detection via Gaussian Curvature Inspired Network,  TGRS

Comments on the Quality of English Language

The quality of the English language in the text is generally good, with a few areas where improvements could be made for clarity and precision. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

Dear authors,m thank you for submitting of your article.

I have some questions and remarks:

 

Figure 1. The overall architecture of LARS   ithis figure is too complicated. It can be devided on 2-3 figures

row 290: and other aerial satellites - aerial images?

Figure 4,5. average detection - Figure 4,5. Average detection

Of the 57 references, only about 4 are outside China. I understand that you are citing locally, but I think a world journal should give an overview of research globally in the introduction section. In general, I think there are many similar articles;I have seen many similar ones. You take some data set generally and free available and put a neural network on it. Adjust some parameters, it comes out a little better, with past approaches, but there is no fundamental evolution. It is clear to me that there is a lot of work in your paper.
If you want to publish this, you need to describe this. Improving a procedure by a few percent just by improving procedures already in use doesn't do much good in science. This is not a negative criticism, just my observation of current trends. Everywhere NN, the design of the papers is similar, the results are similar. Sorry for it.

I don't see too much scientific progress here. Also, describe what this is actually supposed to be used for? For espionage? For tracking people and equipment movement?

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

The adaptive channel attention and large kernel adaptive are interesting solutioon, beacose it is influence on local sharping properties of image.
1. I suggest placing Table 1 after the description of DOTA2.0
2. The datasets are very well-known and do not require detailed description by classes. Perhaps in this part, it should be indicated that the datasets were chosen for comparison with other algorithms; they are good for testing a neural network for multi-assembly or semantic detection. In short, it would be good to justify why these datasets were chosen.

3. The ACA Block involves a combination equivalent to binarization, where the result is concatenated with the original matrix. The effect of this operation is unclear for color. It should only be effective if processed separately by channel. Otherwise, artifacts are possible for objects with intermediate colors (for example, yellow or cian). But further on in the LCA block, we again see color illustrations. It is unclear where the color is separated and processed. Moreover, color  is mentioned on page 6 as semantic features.

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

1. Before introducing the small object detection algorithm, the author should first introduce the specific size range of the small object to be solved in this article;

2. DOTA-v2.0 is not a typical dataset for small object detection, and AI-TOD and SODA-A datasets can be considered for experimental validation;

3. The dataset and evaluation metrics used for small object detection are not representative. It is recommended to refer to the paper "Towards Large Scale Small Object Detection: Survey and Benchmarks".

Comments on the Quality of English Language

Please have the English writing reviewed and approved by the editor.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 4 Report

Comments and Suggestions for Authors

The author has provided a good response to the issues raised in my last review and suggests acceptance.

Comments on the Quality of English Language

Please revise the English writing according to the suggestions of the editor.