GPR Clutter Reflection Noise-Filtering through Singular Value Decomposition in the Bidimensional Spectral Domain

Round 1

Reviewer 1 Report

Dear Authors,

Thank you for the article, it is well structured and presents the results and methods flawlessly.
I am looking forward to further results of your applied research.

Reviewer 2 Report

Review of the manuscript remotesensing-1228994 “GPR clutter reflection noise filtering through singular value decomposition in the bidimensional spectral domain” by Rui Jorge Oliveira, Bento Caldeira, Teresa Teixidó and José Fernando Borges.

 

This is an interesting paper that deals with the use of Singular value decomposition (SVD) to design filters in the frequency domain, useful to remove the clutter reflection noise present in GPR data. The authors have a good dataset, the paper is properly written and well structured, the introductory material is adequate and each section is clearly developed.

 

This manuscript represents a resubmission of a previous version that I had the opportunity to revise. I made some recommendations, mainly related to a reorganization and improvement of the discussion section. I am glad to see that the overall manuscript has been improved, and that the Discussion section in particular has been greatly modified. All the suggestions that I previously proposed has been taken into account, in such a way that now the the pros and cons of the method are clearly exposed,  the limitations of the technique are discussed in detail and it results now more evident the superior results of the method compared with a standard (or classic) GPR data processing. A more detailed comparison with different previous filtering techniques to remove this kind of noise in GPR data has also been developed and included. I consider that the results are adequately supported by the data and, in my opinion, the paper results of interest for potential readers. Consequently, my recommendation is that this new version must be accepted for publication.

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.

Round 1

Reviewer 1 Report

Review of the manuscript remotesensing-994677 “Enhancement of 3D GPR datasets in the spectral domain through singular value decomposition” by Rui Jorge Oliveira, Bento Caldeira, Teresa Teixidó and José Fernando Borges.

 

This is an interesting paper that deals with the use of Singular value decomposition (SVD) to design filters in the frequency domain, useful to remove the background noise present in GPR data. The authors have a good dataset, the paper is properly written and well structured, the introductory material is adequate and each section is clearly developed. The data processing is very nicely described what is not common in the GPR literature. I consider that the results are adequately supported by the data. In my opinion, the paper results of interest for potential readers and consequently it could be accepted after the authors perform a moderate revision:

 

- The only major concern of the manuscript is related to the Discussion section. It mainly consists in a description of the goodness of the results and the effectiveness of the method compared with both the laboratory example and the archaeological excavation results. Moreover, the discussion section repeats a lot of information from the results section that is duplicated again in the conclusions section. However, a discussion is more than this. The discussion must emphasize the pros and cons of the method, not only the positive aspects. What are the limitations of the technique? For example, the filtering results applied to the real radargram (fig. 16d) looks very similar to the input radargram (fig. 16a), so I cannot see in this particular case why the proposed technique improves the results (the same interpretation could be done by just using the input data). However, the improvement in resolution is evident for the case of the slices shown in the 3D case (fig. 17). This kind of observations must be part of the discussion.

 

In addition to this, there exist many examples in the literature dealing with different filtering techniques (e.g. directional total variation minimisation (DTVM) filter, eigenimage processing, multiresolution wavelet analysis, among others) to remove background noise in GPR data, but I have not read any reference to them. At least, a comparison of the present work with similar previous works must be done. Why is this work different? What are the strong points of this technique compared with previous works? What is the novelty of this study?

 

Consequently, the present discussion section must be removed and rewritten according to the aforementioned information.

 

I have a couple of minor comments too:

 

• Lines 118-119: it is stated that ‘The attempt to filter the observed data to bring them closer to the synthetic ones was not possible with commercial software to process GPR data.’ It would be interesting to provide a filtered radargram obtained with a commercial software to confirm that sentence.
• Lines 192-193: ‘the radargram obtained in the laboratory model was considered after applying a custom deconvolution’. Please, explain why deconvolution must be applied before the applying the 2D Fourier transform.
• Lines 357-358: ‘These principal components, weighing less than the first component, sufficiently represent the data, and their deletion does not compromise the structural integrity of the dataset’. I do not understand the meaning of the sentence. If the principal components, except the first one, sufficiently represent the data, their deletion would really compromise the dataset, because if they are deleted the only remaining information would be the noise component.

 

Regarding the figures, in general they are of good quality but some minor modifications must be carried out:

 

- Figures 8 and 14: the label of the y-axis is incorrect (component wight (%)) it must be replaced by component weight (%).

- There are two different figure 17 captions. The second one (at line 461) must be changed to figure 18.

 

Hope these comments might help.

 

Regards,

Reviewer 2 Report

Many thanks for this well structured article. The method is excellently described in detail and clearly understandable.

The results look very promising and I am sure that the formula you presented will be well appreciated in the scientific community.

Reviewer 3 Report

There is no need to explain PCA and SVD in such details. These methods are already well known and very documented. 

A review of the free available processing tools for GPR data should be provided and performance results should be compared.