TSVR-Net: An End-to-End Ground-Penetrating Radar Images Registration and Location Network
Round 1
Reviewer 1 Report
Figure 1. Letters are too small
Figure 3 Should be bigger. Hard to see details
Figure 10. Really poor figure. No scale bar, no coordinate grid.
Figure 15. All GPR profiles must contain information regarding time axis and horizontal axis (distance)
Line 101. EM waves are not created below the surface. GPR receiver receivs reflections of the transmitted EM impulses.
Line 102. not only dielectric permittivity. via reflection we can receive impulses with different amplitudes and sometimes slightly different frequencies. I do not understand about what other characteristics the authors are speaking?
line 121. It is not advaisable to use term "echo" while speaking about GPR data. It creates wrong impression that GPR has something to do with SONAR. Better to use term GPR image or some other common term used in GPR literature.
LIne 138. It must be specified what authores mean by low and high frequency. AS well as there should be additional information regarding used GPR system. Its frequency etc.. Because those are crutial questions. There are numerous studies that show significant difference between data obtained via different GPR antennas etc..
Figure 7. It is doubtfull that current GPR data shows reflection form a boundary. Most likely this is some background noise. At least if no bigger data peace is shown, it seems so. Also information regarding scale of horizontal and vertical axis is necessary. GPR data typically contains a lot of noise and unnecessary signals that have nothing to do with subsurface features. In current view it is impossible to say anything about presented data.
line 193. It is interesting that authors choose LP filter to enhance the data. Usually background "noise" has low frequency in typical GPR data. So there is slight probability that authors are tracking not only valuable reflections that contain information about subsurface but also random noise data.
line 350. So authors propose that rainwater can pour through pavement???
Section of conclusions is more like description on what has been done and not like general conclusions about problem authors tryed to solve. This section must be rewritten.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Review of the manuscript "TSVR-Net: An End-to-End Ground-Penetrating Radar Images Registration and Location Network", by Bi et al.
This paper introduces a novel localization technique based on ground-penetrating radar method (GPR). The authors propose an end-to-end TSVR-Net-based regression localization method that effectively provides the stability of the localization system with high accuracy. The paper is well-structured and logical, and it is a worthy paper for publication. The paper provide a new idea for achieving a stable and reliable localization system using the GPR. However, after I read the whole article, I have some questions?
1. What are the advantages and disadvantages of GPR as a localization system for autonomous driving?
2. What is the difference between the GPR-based method and the traditional LIDAR or camara-based localization system, and what is the comparison between their effectiveness for localization?
3. We know that GPR generally emits signals to the subsurface and detects the buried objects in the shallow surface, what is the connection between the information of the subsurface structure obtained by the GPR and the information on the road?
4. What is the response time for automatic determination of subsurface echo signals in the GPR images? Can this be compared to the speed of a car driving at high speed and can it react? How does the machine automatically identify and locate in a short time?
5.How effective is the ground-penetrating radar-based localization system in practice? What are the criteria for judging?
6. What are the echo characteristics of the reflected echoes below the ground surface acquired by GPR and used for localization?
7. How likely is the method proposed by the authors to be used for future autonomous driving? How reliable is it? A brief description by the authors is needed.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
Fig. 1: Y- the y scale for the C-Scan can't be 'travel time' but should be distance along travel direction.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
No further comments
Reviewer 2 Report
The authors responded to all my comments and, I recommended the paper to publish in remote sensing.
