A Gunshot Recognition Method Based on Multi-Scale Spectrum Shift Module
Round 1
Reviewer 1 Report
The paper proposes a neural network (NN) based on a multi-scale spectrum shift module to target the gunshot detection problem. The proposed NN can realize the interaction among spectrum information of gunshots through the shift module to benefit mining the relevant information among the gunshot spectrums. Also, the method can avoid information loss during the down-sampling process and resolving the problem of low SNR and susceptibility to interference. The evaluations demonstrate the performance of the proposed NN.
Overall, the paper is interesting and targets a meaningful problem, and the proposed solution works.
There are a few concerns below and it is advised to address them
1. The motivation is not strong enough. The paper states "All the above gunshot recognition methods are of some limitation". It is not very clear what the limitations are. It is advised to illustrate them clearly.
2. Regarding the spectrum shift module, the channel is moved up/down by 1/2 bits to mine more information. Actually, this method is not very convincing. As the adjacent channels are typically correlated with each other (i.e., the information between adjacent channels correlates strongly), with the moving operations, the adjacent channels may contain uncorrelated information eventually. In this case, why does the convolution operation benefit from the "messy" channels?
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The authors describe the problems of using video to detect gunshots in city streets and then explain why the use of audio detection has some superior detection properties. A full explanation is then given of their proposal for a small neural network based on multi-scale Spectrum Shift principles.
The next section outlines their detailed proposal for such a fully interconnected neural network.
Their development of their innovations is well presented and results are well presented together with a well-developed analysis of the system.
The authors’ coverage of the problems with current datasets is discussed and their approach to these problems.
The results show an improvement in a number of fields to the problem of “gunshot” detection and is a contribution to the field of research especially in the use of frequency shifting.
The paper is well written, clear and well laid out.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
The reviewer comments are attached as pdf document
Comments for author File: Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
