A Cascade Network for Pattern Recognition Based on Radar Signal Characteristics in Noisy Environments
Round 1
Reviewer 1 Report
The content presented in "Section 3. Radar Signal Detection in Noisy Enviroment" is classic so it's not necessary to decribe it in detail.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This paper proposes a neural-network-based pattern recognition method for radar signals under noisy environment. Several points should be modified.
1. In the title, it is “noise environment” while in the body article it is “noisy environment” which one is correct?
2. There are multiple hyper parameters in the proposed method. It is not clear how the authors justified them. (E.g., in Fig. 6, this reviewer finds dozens of parameters.).
3. In the experimental results, the authors applied conventional methods which parameters are apparently not optimized.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Authors study the radar target recognition problem from an AI perspective. Radar is an extensively studied subject, and authors provide a limited literature review. The proposed approach is clearly explained, and lack of benchmark datasets is clearly mentioned. Authors create a reasonably detailed dataset and demonstrate that the proposed approach does work.
A comparison with traditional non-neural network based radar target recognition methods would improve the quality of the paper.
Authors do not study different SNR levels, and how the detection performance will degrade with increase noise.
-
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
In this paper, the authors propose a cascade network that includes a multi-scale convolutional attention network (MSCANet), for pattern recognition based on radar signal characteristics in a noise environment. The paper has a good idea and, generally speaking, the theoretical background of the paper is written in a clear and consistent way. However, the paper contains several shortcomings that should be corrected in order for the paper to be published in the journal. The mentioned shortcomings and recommendations for their elimination are as follows:
1. 1.It is necessary to explain the recognition network training based on MSCAnet in more detail. Apart from the analytical form of the cost function being optimized as well as the type of regularization applied, it is not clear which optimization method was applied, whether the mini-batch approach in training was applied or not, how many training iterations were applied and other training details.
2. 2. The section dedicated to traditional machine learning algorithms lacks details on the applied training methods for each individual approach listed. Also regarding the MLP approach, the authors state that they used MLP neural networks with a maximum of 1000 neurons in the network. It was not stated how many layers the MLP network contained and what the distribution of neurons was across layers.
3. 3.In Table 3, which refers to the recognition accuracy of several networks in different environments, the Process time value in seconds is given. It is not clear on which reference hardware platform those values were achieved.
4. 4. The paper uses a lot of abbreviations (especially in the pictures) that are not explained (for example: PA, PW, DOA TOA,...). In order for readers who are not in the field of radar technology to understand the proposed method more easily, the authors are recommended to explain each abbreviation in the text of the paper.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 4 Report
The authors satisfactorily responded to the reviewers' comments and improved the paper. The paper is recommended for publication in the "Remote Sensing" journal.
Author Response
Thank you very much for your contribution to the improvement of our paper. Your comments will also provide guidance for our subsequent research.
