Embedding Soft Thresholding Function into Deep Learning Models for Noisy Radar Emitter Signal Recognition
Round 1
Reviewer 1 Report
This paper presents a Embedding Soft Thresholding Function into Deep Learning Models for Noisy Radar Emitter Signal Recognition. The authors propose to a soft thresholding function is embedded into deep learning network models as a novel nonlinear activation function. The proposed method is characterized by flexible nonlinear conversion and ability to obtain more discriminative features compared with conventional activation functions..
In section I - Introduction: the authors present the scope of the proposed work and present some related SoA papers.
In section II – Methodology: the authors present their approach to achieve their goals, where the concept of SofT module presented in the embedded deep learning methods will improve the recognition accuracy of noisy radar emitter signals. For this achievements, the authors propose a soft thresholding function and and the corresponding Soft thresholding module.
In section III – Experimental and results, the authors present five typical activation functions and compare them in this work.In order to verify the effectiveness of SofT the authors generate seven representative modulation types of radar signals. Two neural networks, ConvNet and ResNet, were used to compare the performance of different activation functions for radar emitter signal recognition. A nonlinear reduction method was used, namely t-distributed stochastic neighbor embedding was used to analyze the high-dimensional features learned by the network models.
Overall the work is interesting and has a scientific soundness, namely the presented concept. The presented SoA is very focused and the works are recent. The obtained results are presented with clearness and are compared under Gaussian and Laplacian noise, in order to obtain the recognition accuracy of different activation function.
The authors must renumber the Conclusions section, that is presented in 3.5, but should be setcion 4. The authors should consider present some topics for future work.
In line 185 the authors must include the missing reference.
Some text formatting and English sentences must be reviewed.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
The authors proposed a soft thresholding function (SofT) that was embedded into deep learning network models as a novel nonlinear activation function to achieve accurate radar emitter signal recognition results. The overall work is fine. I have the following minor comments.
1) The authors need to revise the abstract to demonstrate how the new function outperforms the others. In the abstract, some numerical conclusions are required.
2) In figures 5 and 6, please provide sentences demonstrating what various colors mean.
3) Please split figure 7 into multiple figures to better demonstrate the results.
4) Write a statement in conclusion to reflect the drawbacks of the proposed function.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf