GMT-WGAN: An Adversarial Sample Expansion Method for Ground Moving Targets Classification

Round 1

Reviewer 1 Report

The paper is interesting and well written.

I am satisfied with the scientific details of  the version I have read. The paper is very well written, explained very well with many relevant references and presents very good results. The authors seemed to be very familiar with the area which could be very easily seen via the various references they cited. 

Author Response

Thanks a lot for your recognition and support for our work! We will continue to conduct in-depth research in this field and make good research findings.

Reviewer 2 Report

In this paper, the authors tackled the problem of classifying ground moving targets using deep learning. Specifically, the authors designed a generative adversarial network-based technique to synthesize images of ground moving objects that are later fed to a convolutional neural network for classification. The ideas presented in the manuscript are sound, but the paper would greatly benefit from addressing the following issues:

1. The authors should carefully revise the language (perhaps with a help from a native speaking colleague), as there are grammar issues and odd statements (even in the abstract). Additionally, please revise the way of referencing papers (e.g., “[12] improved the model in MOCAP…”).
2. I enjoyed seeing a very detailed discussion of the architectures suggested by the authors in this work – it would be great if the authors could make their implementation publicly available which would make the experiments fully reproducible very easily.
3. The experiments show that there is indeed something promising going on, but the comparison with other data augmentation methods is a bit limited – I suggest expanding it to show the advantages of the introduced technique. Also, it would be useful to see the results of the statistical analysis (backed up with appropriate statistical testing) to see if the differences across the results are statistically significant.

Author Response

In this paper, the authors tackled the problem of classifying ground moving targets using deep learning. Specifically, the authors designed a generative adversarial network-based technique to synthesize images of ground moving objects that are later fed to a convolutional neural network for classification. The ideas presented in the manuscript are sound, but the paper would greatly benefit from addressing the following issues:

We appreciate your recognition and affirmation of our work. Meanwhile, we have carefully revised the improper parts in the paper.

 

1.The authors should carefully revise the language (perhaps with a help from a native speaking colleague), as there are grammar issues and odd statements (even in the abstract). Additionally, please revise the way of referencing papers (e.g., “[12] improved the model in MOCAP…”).

We have carefully checked and corrected the errors invited the professional English native speakers to revise the English in this paper. We look forward to your review again!

 

2.I enjoyed seeing a very detailed discussion of the architectures suggested by the authors in this work – it would be great if the authors could make their implementation publicly available which would make the experiments fully reproducible very easily.

We are very sorry for that our implementation is not suitable for opening. If you need it, you can contact us by email.

 

3.The experiments show that there is indeed something promising going on, but the comparison with other data augmentation methods is a bit limited – I suggest expanding it to show the advantages of the introduced technique. Also, it would be useful to see the results of the statistical analysis (backed up with appropriate statistical testing) to see if the differences across the results are statistically significant.

Your suggestion is very useful to us! We expanded our comparative experiments and calculated the statistical analysis results, which were added to the new manuscript and highlighted in yellow.

Author Response File: Author Response.docx

Reviewer 3 Report

I find myself in a very difficult situation here.  The technical content of this paper is good enough.  I have no major problems here.  However, the quality of writing, that is the quality of English in the main, is really poor.  I do not think that papers as poorly written as this can be accepted by a reputable journal and I doubt that the authors can improve it sufficiently by themselves - hence my recommendation.  They would need a professional and somebody who speaks good English.

Just some (n.b. a tiny number compared to the total) examples:

• "submerged in clutter" - "submerged" is a wrong word.
  
  
• "challenge; Moreover" - repeated improper use of semi-colon where a full stop should have been used.
  
  
• "capacity; Besides" - same as previous.
  
  
• "Besides, " - repeated use of "besides" not in the spirit of English.
• "by feeding augmented samples...the classification performance of ground moving targets has been improved." - grammar all over the place.
  
  
• "on various datasets" - poor word choice ("various")
  
  
• "which is an issue" - grammatical number mismatch.
  
  
• "It is well known that Doppler effect" - don't state what is well known, just state what you want to state ("The Doppler effect...").
  
  
• "The pedestrian movement is a highly coordinated non-rigid movement of the brain, muscles, nerves, joints and bones." - a terrible statement, sorry!
  
  
• "recognition, [15] classified pedestrians" - references are not part of text; this reads as "recognition, classified pedestrians"; you should write "recognition, Garreau et al. [15] classified pedestrians".
  
  
• "As the most commonly used time-frequency analysis method" - how can you say that? Where is your evidence? Has somebody actually examined this? Please do not write statements like this.
  
  
• "STFT has its advantages, that is, it has not considered the effect of cross terms," - again, all over the place grammatically and otherwise.
  
  
• "un-armed" - this is just a simple, single word: "unarmed"
  
  
• "Figure 7 is time-frequency spectrograms" - grammatical number mismatch again.

And so on, and so on, for many many more examples.

The authors should also cite some other recent work on WGANs which also argue for and show their superiority, e.g. Magister's "Generative image inpainting for retinal images using generative adversarial networks" (2021).

Author Response

I find myself in a very difficult situation here. The technical content of this paper is good enough.  I have no major problems here.  However, the quality of writing, that is the quality of English in the main, is really poor.  I do not think that papers as poorly written as this can be accepted by a reputable journal and I doubt that the authors can improve it sufficiently by themselves - hence my recommendation.  They would need a professional and somebody who speaks good English.

Thank you very much for your advice! We grateful for your patience, and your guidance is great help to our future work! We have carefully checked and corrected the errors, and we have invited the professional English native speakers to revise the English in this paper.

 

The authors should also cite some other recent work on WGANs which also argue for and show their superiority, e.g. Magister's "Generative image inpainting for retinal images using generative adversarial networks" (2021).

Some recent work on WGANs are cited in the new manuscript and highlighted in yellow.

In [36], the WGAN with a semantic image inpainting algorithm was used to devise more realistic retinal images for medical teaching purposes. In [37], a novel framework by concatenating a super resolution GAN and a WGAN was proposed to increase the performance of a backbone detection model.

1. Magister, L. C.; Arandjelović, O. Generative Image Inpainting for Retinal Images using Generative Adversarial Networks. Annual International Conference of the IEEE Engineering in Medicine & Biology Society, Mexico, 2021.
2. Fu, Y.; Zheng, C.; Yuan, L.; Chen, H.; Nie, J. Small Object Detection in Complex Large Scale Spatial Image by Concatenating SRGAN and Multi-Task WGAN. International Conference on Big Data Computing and Communication, Deqing, China, 2021.

 

We look forward to your review again!

Round 2

Reviewer 3 Report

I appreciate the authors' constructive receipt of my advice.  The manuscript is indeed improved. there is one thing that should be remedied before publication though.  Namely, please note that references are not part of text - they are markers which are not read. In other words, "In [11], a few..." reads *exactly the same* as  "In, a few...", i.e. it does not make any sense.  Rather, it should always be something like this instead: "In the work of Smith et al. [11], a few..."