Automatic Detection of Diseases in Tunnel Linings Based on a Convolution Neural Network and a Support Vector Machine
Round 1
Reviewer 1 Report
In this paper the authors are introducing a method to automatic detection of diseases in tunnel lining utilizing ground penetrating radar.
The paper is well written, introduces a solid theoretical background, and offers good results and a thorough comparison to other approaches. However, I suggest to the authors to improve the english in the manuscript and to increase the number of references.
Also, please, replace the figure 6 with english words.
Author Response
Thank you sincerely for your valuable comments and helpful suggestions. Your comments are helpful for improving our paper and have important guiding significance to our researches. In the revised manuscript, the English has been improved and we have tried our best to make the issues clear. We hope that all the comments have been satisfactorily addressed. If some points are still not expressed clearly, we hope that we have opportunity to continue to revise. In the following, the item-by-item responses are given.
Author Response File: Author Response.docx
Reviewer 2 Report
I would like to thank the authors for their work.
This article concerns the implementation of a procedure for the automatic detection of tunnel lining defects.
1) I think the summary needs to be improved and not just be a list of techniques used.
2) Authors should better specify the configuration(s) of the gpr (ground coupled, air coupled, impulse, central frequency...)
3) We need more precision for the gpr data model and the Kalman filter.
It is not clear for me how you apply the kalman filter with the two hypothesis. The use at the end of a pseudo code would help understanding.
4) The use of the gprMax software without citing the authors is aberrant.
5) The description of the database is partial. All the parameters of the simulation must be given. Simulation parameters are not results.
6) Figures 4 and 5 need to be describe.
7) Figure 6 is not understandable.
8) 2.3 is detection or classification? Why restrict the database in 2.3?
9) How do you choose your features ?
10) The parameters of SVM?
11) How do you define the signal to clutter ratio?
12) Figure 11 need more description. Definition of the moving average, Wavelet transform. Difference between SCR nd SNR ?
13) Depth is in meter when you are in ns it is time.
14) Use traditional metrics for the evaluation of the accuracy of SVM, like confusion matrix for example.
15) the conclusion needs to be rewritten
Author Response
General Response:
Thank you sincerely for your valuable comments and helpful suggestions. Your comments are helpful for improving our paper and have important guiding significance to our researches. In the revised manuscript, we have tried our best to make the issues clear. And we hope that all the comments have been satisfactorily addressed. If some points are still not expressed clearly, we hope that we have opportunity to continue to revise. In the following, the item-by-item responses are given.
Author Response File: Author Response.docx