Extraction of Micro-Doppler Feature Using LMD Algorithm Combined Supplement Feature for UAVs and Birds Classification
Round 1
Reviewer 1 Report
Dear authors,
I have a few comments:
- Describe the radar before the results you talk about (before figure 1)
- Table 1 should be extended (Pulse length, Blind zone, if you are using any modulation of the pulse)
- Fig 6, what are the purple curves?
- Fig 1. what if there is more drones (group of drones)?
- Lines 167-170: makes the text revision in connection to the figure 1(Fig 1.b - Target close to radar, but in text is: Fig 1b target is far away.)
- Figure 3 seams that you used for the spectrum different times from the waterfall.
- You write about the smooth trajectory of the drones in the comparison with the birds, what if the dron imitates the birds trajectory, did you try if the UAV can do it? And if it can be recognised the target after that?
Wish you a nice day.
Author Response
Dear Reviewers：
We are grateful to the Reviewers for taking the time to review our manuscript and providing us with their valuable comments. There is no doubt that these comments are valuable and very helpful for revising and improving our manuscript. Our point-by-point reply to your comments, please refer to the uploaded PDF document for details
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The authors address RADAR micro-Doppler processing for separating the signal reflected by UAS with respect to birds. Feature extraction is discussed in detail in assessing body cross section, reflected energy and separating the micro-Doppler features expected to clearly separate the UAS signature from birds.
The topic of micro-Doppler feature extraction and classification of UAS signatures has already been widely investigated in the literature so that the contribution of the authors is a declination of existing work. The authors compare with Empirical Mode Decomposition (EMD) which seems to be the reference method in their work. Beyond the fact that the contribution appears as merging multiple estimators to try and improve the detection rate, with each estimator well known from the literature and hence with little scientific novelty from my reading, I have a couple of concerns about the underlying soundness of the analysis:
- in two places, arbitrary thresholds are used ("0.08 is the empirical value" on l.221 and "0.05 is set empirically" on line 283) which significantly weaken the argument. Why not used a Constant False Alarm Rate (CFAR) adjustable threshold? This threshold seems rather trivial considering the complex signal processing involved later, e.g. 2.2.4 with the Random Forest Classification.
- are the results significantly better? No error bar or uncertainty estimate is given. The 2.56% (two decimals on a couple of percent improvement) seems either negligible, or I would like at least to see how the relevance of this difference is assessed. Table 5 provides two decimals on classification success rate (and l.450 only provides 99.1% with a single decimal), but how is the errorbar established and the number of relevant displayed decimals assessed? Same for table 7.
On the layout of the document, many English grammar mistakes seem to be found in the manuscript. Despite English not being my native language, I believe the following are incorrect:
l.9: "be revealed clearer" sounds awkward
l.53: is RGB "Red Green Blue"? I fail to understand why image processing is included in a manuscript about RADAR reflection processing
l.57: "Separating the rotating parts will enable them to run" sounds incorrect. "will allow their signature to be identified without intereference"?
l.60: does "T-F" refer to Time and Frequency?
l.75: "While the radar observation target can also observe" sounds incorrect
l.77: that extracting -> that extracts?
l.80: "staring radar" is not defined. I am not familiar with this RADAR architecture and its definition might be useful to the reader.
l.98: and completed -> and complete
l.99: introduced -> introduces
l.106: "time frequency" should be introduced earlier, at the first occurrence of the T-F acronym, which can then be used here
l.108, 114, 116, 136, 180, 181: returns signal (many times) -> return signals or returned signals
l.124: the most common being taking (remove "one is" I believe)
Fig.1 caption: is the description of (a) and (b) correct? It looks to be like (a) with its strong spectral features is collected close to the RADAR and (b) with its weak features has been collected far from the RADAR
l.149: to have the largest -> exhibit the largest?
l.152: "mixed together with" I think I understand the meaning but the spelling sounds awkward
l.157: "PC material" what is PC (since the acronym is used later for Personal Computer, I am not familiar with PC material)
ll.157-162: beyond the sentence sounding awkward ("can be observed"), the statement is rather trivial and obvious.
l.167: noise. So that -> noise, so that
l.168: "are shown that the condition" sounds incorrect
l.170: "gliding state cannot observe" sounds incorrect, a state cannot observe
l.184: target passes more volatile sound incorrect
l.198: is generated -> remove "is"
l.203: the first occurence of "SVD" is not define while l.205 explicits "singular value"
l.216: "can obtain the target signal" sounds incorrect
l.218: singular values are (not "is")
l.234: And then -> no verb in this sentence, maybe merge with the previous one
l.235, 439, 447: multiple times two dots after a figure number (Figure 3.., 7.., 8..)
l.254: are shown
l.255: separates
l.264 and 281: Which -> with
l.371: proposed method is compared
l.373: stating that the test was performed using Matlab weakens the executing time statement given later (Tables 3 and 4) since the interpreted language is hardly designed for speed. A 2.6 GHz CPU hardly provides any information without the CPU model and possible SIMD instruction set. I am not even sure the relative time is relevant since Matlab is strongly flawed with loops and excels at linear algebra matrix implementation, so the validity of the comparison is unclear to me.
l.404: species varying with time
l.428: classication -> classifying
l.478: why is log-Fourier relevant? (as opposed to a Fourier transform)? what does the logarithm bring?
Author Response
Dear Reviewers：
We are grateful to the Reviewers for taking the time to review our manuscript and providing us with their valuable comments. There is no doubt that these comments are valuable and very helpful for revising and improving our manuscript. Our point-by-point reply to your comments, please refer to the uploaded PDF document for details.
Author Response File: Author Response.pdf
Reviewer 3 Report
The authors propose a local mean decomposition technique to distinguish between the micro-Doppler signatures of UAVs and birds in radar automatic target recognition applications. The authors specifically target the limitations of m-DS detection and propose an algorithm that better utilizes the information available in a spectrogram. They demonstrate 2.56% reduction on equal error rate on a benchmark dataset.
[44] - Some of the citations appear to be out of order. I do not know what standard this journal uses, but I usually see the references listed in the order they appear in the text (i.e. you cite [2], then [24], then [3-8], I would expect this to read [2], [3], [4-9]). If this doesn't apply to this journal, please ignore.
[103] - The introduction is well written and clearly communicates the context, the specific problem statement, and how your proposed solution addresses. Aside from some minor grammar issues, I have no further suggestions for this section. Nicely done.
[128] - There are some very minor language issues on this page but it is otherwise clear.
[129] - The axis labels on the subfigures in Figure 1 could be slightly larger and maybe bold, otherwise the figure is good.
[158] - What is "PC material"? Did you mean PVC or something else? If so, please consider spelling it out the first time you introduce it.
[162] - The formatting of this page is a little clumsy and I think the bullet points are unnecessary. Perhaps consider instead using bold text followed by a dash for the headers instead of a bulleted list.
[187] - I would like to see a couple of sentences in the caption of Figure 2 explaining the flow of the depicted classification system. The diagram is good, but it can be even better with some supporting text.
[233] - The axis and tick labels on Figure 3 need to be much larger, and since you're already using the whole page you can write a lot more in the caption.
[263] - The formatting makes this difficult to read, consider using bold text when you write "Feature#:".
[272] - Figure 4 also needs larger labels and more supporting text in the caption.
[301] - The table of operating parameters and images of the experimental setup are excellent, thanks for including these.
[423] - Do you have any insight as to why the LMD algorithm executes so much faster? Even if the algorithms performed the same, an order of magnitude reduction in the processing time alone would be a significant result. Can you add some explanation as to why this may be?
[464] - Figures 7 and 9 need larger labels.
[480] - This paragraph was an excellent explanation on why the different state of the art techniques perform differently, thanks for including this.
Thank you for your patience while I prepared this review, and thank you for your contribution to the journal.
Author Response
Dear Reviewers：
We are grateful to the Reviewers for taking the time to review our manuscript and providing us with their valuable comments. There is no doubt that these comments are valuable and very helpful for revising and improving our manuscript. Our point-by-point reply to your comments, please refer to the uploaded PDF document for details
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Dear authors,
thank you, that you reflected my coments, but please include also the following points:
Please, write in the conclusion, why you didn't try use the group of the drones and that it is your plane for the future.
Please add the respond for the point 7 (you sent me) also inside the text of your paper.
Wish you a good luck with your next research.
Author Response
Dear Reviewers：
We are grateful to the Reviewers for taking the time to review our manuscript and providing us with their valuable comments. There is no doubt that these comments are valuable and very helpful for revising and improving our manuscript. Our point-by-point response appears below, in which we first echo the comments and then a response to them.
Author Response File: Author Response.pdf
Reviewer 2 Report
The authors have answered my comments satisfactorily and I believe the paper can now be published. Thank you.
Author Response
Dear Reviewers：
We thank the Reviewer for the comment. Your affirmation will motivate me to improve.
Author Response File: Author Response.pdf
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
