A Two-Stage Track-before-Detect Method for Non-Cooperative Bistatic Radar Based on Deep Learning
Round 1
Reviewer 1 Report
Comment to remotesensing-2474401
This paper combines the deep learning algorithm to propose a two-stage track-before-detect (TBD) method for tracking low-SNR targets. This paper addresses the very challenging problem of tracking weak targets in the presence of clutter, misdetection and data association uncertainty. The also presented numerical studies to validate the proposed algorithms. The experiment results demonstrate that the proposed method can achieve multi-target tracking performance in the challenging scenario with highly clutter. Overall, this is an excellent paper. The paper is well organized, and very well written. It is a pleasure to read. Hence, I believe that the manuscript deserves publication. I only have some minor comments which I hope the authors can revise the manuscript to improve the standing of their work.
1. Generally, traditional model TBD method is computationally efficient implementation for tracking low-SNR targets compared with data-driven version, such as deep learning method. Please author explains why the deep-learning TBD consumes less time than model version.
2. The deep learning algorithm should be more important for the design of the whole method. Therefore, author should introduce the deep learning algorithm in detail in this manuscript.
3. The TBD method based on low threshold has been discussed and applied in many references [1,2]. Why does author claim that the low-threshold TBD method is firstly used for extracting the targets.
4. The readability of this manuscript needs further improvement because there are a few typos or grammar mistakes in the manuscript. The authors must proofread the whole paper for improving English writing in order to get accepted.
This article presents deep learning-based TBD algorithm for tracking multiple weak targets. Overall, it is suggested to be accepted with major revisions for this manuscript.
Reference
[1] Chenghu Cao, Yongbo Zhao, Xiaojiao Pang, Zhiling Suo, Sheng Chen. An efficient implementation of multiple weak targets tracking filter with labeled random finite sets for marine radar. Digital Signal Processing, 101(102710), 2020.
[1] Wei Yi, Zicheng Fang, Wujun Li, Reza Hoseinnezhad. Multi-frame track-before-detect algorithm for maneuvering target tracking [J]. IEEE Transactions on Vehicular Technology, 2020, 69(4):4104-4118.
[2] Jinghe Wang, Wei Yi. An efficient recursive multi-frame track-before-detect algorithm [J]. IEEE Transactions on Aerospace and Electronic Systems, 2018, 54(1):190-204.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
In this manuscript, the authors propose a novel track-before-detect method of deep learning for non-cooperative bistatic radar. It is a topic of interest in related areas. The information about the tracks is obtained by the networks of deep learning. However, there are some problems which need to be solved before it is considered for publication.
My comments are:
(1) Research background of deep learning in track-before-detect needs to be supplemented in the introduction.
(2) In general, the English grammar and sentence structure needs improvement.
(3) As shown in Figure 5, there is a LSTM network in the network structure, which also has related formulas in page 10. Thus, A more detailed structure of LSTM should be augmented.
(4) In the experiment, the time cost about training of the network needs to be supplemented. Besides, the setting of false alarms is relatively low. Whether the increasing number of false alarms has effect on the results of experiments or not? It should be discussed to confirm the generalization ability of the network.
In general, the English grammar and sentence structure needs improvement.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
Thank you for your creative study on this topic.
1. Line 48 & 49, the first time you use DBT & TBD, you’d better to show the full name.
2. Line 48, “the traditional DBT method finds reliable detection and stable tracking of the target difficult.” This sentence is not well clarified or referenced.
3. Have the authors considered multipath interference? In complex maritime environments, there is the problem of multipath propagation. This means that signals may propagate through multiple paths to reach the target and the receiving antenna, resulting in multipath interference in the echo signal, which in turn affects the accuracy of target tracking.
4. Section 3.1, why the order of the length is not consistent with the order of iteration rounds?
5. The tile of section 3.3 is Complexity analysis. Please consider the accuracy of this title in the context of this section. This section is closer to real-time performance analysis.
6. The discussion section does not address the previous studies and conclusions comprehensively, would authors like to enrich the content please.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
I have no questions.
Author Response
Dear reviewer:
Thank you for your decision and constructive comments on my manuscript.
Reviewer 2 Report
In this manuscript, the authors propose a novel track-before-detect method of deep learning for non-cooperative bistatic radar. It is a topic of interest in related areas. The information about the tracks is obtained by the networks of deep learning. The authors have made corresponding modifications to the original manuscript according to the comments. However, there is a problem which need to be solved before it is considered for publication. Please note that one symbol had better have one meaning. There is a symbol(letter L) in the manuscript that represents both baseline distance and system loss.
Minor editing of English language required.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
