Multi-Frequency GPR Data Fusion with Genetic Algorithms for Archaeological Prospection
Round 1
Reviewer 1 Report
this is a very nice paper. I am happy to see all the math modeling for doing what you have done. I am not familiar with exactly what you are doing, but have worked with the Norwegian multiple frequency system and know how they do things. There they are using various frequencies to study various depths. I am not sure you are doing this, and can't find where you outright say that. Am I wrong? That might be a very important way to present to the readers how important this method is, if that is where you are going with it.
The profiles are WAY too colorful for me also. Perhaps just gray scale with annotations showing what features in the ground you care about would be enough?
Author Response
Comments and Answers
(1) This is a very nice paper. I am happy to see all the math modeling for doing what you have done. I am not familiar with exactly what you are doing, but have worked with the Norwegian multiple frequency system and know how they do things. There they are using various frequencies to study various depths. I am not sure you are doing this, and can't find where you outright say that. Am I wrong? That might be a very important way to present to the readers how important this method is, if that is where you are going with it.
Thank you so much for your suggestions. You are right, various frequencies for GPR are used to study various depths. GPR data fusion approaches are proposed to at least partly overcome the unavoidable trade-off between penetration and resolution, while improving the imaging of the subsurface. So in the Introduction, we describe that “According to the propagation characteristics of the EM wave, higher frequency GPR signals offer higher resolution while lower frequency GPR signals offer greater depth of penetration. In order to avoid limitations in imaging, multi-frequency GPR systems are often used for complicated subsurface archaeological conditions” (Lines 29-33), and “GPR data fusion approaches, automatically or semi-automatically integrating data measurements extracted from antennas of different frequency and combining them into a single representation, are therefore proposed to overcome the unavoidable trade-off between penetration and resolution, and to produce accurate, robust, and re-liable estimates of different subsurface structures” (Lines 39-43).
(2) The profiles are WAY too colorful for me also. Perhaps just gray scale with annotations showing what features in the ground you care about would be enough?
Thank you so much for your suggestions. Gray scale are really often used in GPR, but we also used two very common colormaps, both of which are welcome in GPR community, and we believe they can help us to identify the potential archaeological structures.
Reviewer 2 Report
The paper proposes a merging of low-frequency and higher frequency GPR data through a suitable weighting coefficient calculated with a genetic algorithm. Experimental data are processed from two different cases studies in China and in Italy, showing that the joining of the data offers the possibility of a larger penetration and a better resolution at least of the shallower targets, and at the same time does not show any “seam” due to the joining of different data.
The paper has points of interest, but some things should be better clarified. In particular:
- The flux diagram in figure 1 is not clear. In a flux diagram the branch departing from any box should be unique or conditionally different. Instead, e.g. from the box “resampling” there are two arrows and no explanation is provided about the flux that the signal should follow and why.
- The sentences in lines 110-113 should be expanded: the accumulated energy by what entity? And “accumulated” with respect to what? Does it means accumulated over the time?
- At lines 170-171, it is specified that the topographic correction is done with an average propagation velocity of 0.12 m/ns. Is the topographic correction correlated to the propagation velocity? I woud say yes if also a time-depth correction is performed, but this is not the case of fig. 7. Of course the propagation velocity is important and it is correct to report it, but it is not directly related to the topographic correction if this is done in time domain.
Further points are the following:
- There is a larger literature about the joining of different frequency data, and there are also methods that join the reconstructions rather than the data. So, some more references and some comparison between different methods would be useful.
- At line 130, I would say “single band” rather than “single frequency”, that might be misleading.
Author Response
Comments and Answers
(1) The flux diagram in figure 1 is not clear. In a flux diagram the branch departing from any box should be unique or conditionally different. Instead, e.g. from the box “resampling” there are two arrows and no explanation is provided about the flux that the signal should follow and why.
Thank you so much for your suggestions. You are right, in a flux diagram the branch departing from any box should be unique. But we have to say that sometimes one output is derived from multiple inputs. The flow chart is just a schematic diagram, which is used to understand the text description, although we think we have described related contents well in the text. About the “resampling”, there are two arrows, because one branch is used to calculate the weight factor, also by virtues of genetic algorithms, and the other branch is used to generate the final fused result, also combined with the weight factor.
(2) The sentences in lines 110-113 should be expanded: the accumulated energy by what entity? And “accumulated” with respect to what? Does it means accumulated over the time?
We added the content as suggested and changed it as “where n0 means the sample where the accumulated energy over the time reaches a threshold value with regard to the signal from the high frequency GPR profile.”
(3) At lines 170-171, it is specified that the topographic correction is done with an average propagation velocity of 0.12 m/ns. Is the topographic correction correlated to the propagation velocity? I woud say yes if also a time-depth correction is performed, but this is not the case of fig. 7. Of course the propagation velocity is important and it is correct to report it, but it is not directly related to the topographic correction if this is done in time domain.
We used an average propagation velocity of 0.12 m/ns to perform the time-depth correction in Fig. 7, of course, in order to keep unity with preliminary publication (Zhao, W.; Forte, E.; Levi, S. T.; Pipan, M.; Tian, G. J ARCHAEOL SCI 2015) [1], the 20 ns corresponds to 61.6 m on elevation. In the vertical axis, we can see that (200-20 ns)/2*0.12m/ns=61.6-50.8 m.
(4) There is a larger literature about the joining of different frequency data, and there are also methods that join the reconstructions rather than the data. So, some more references and some comparison between different methods would be useful.
Thank you so much for your suggestions. The main idea of our paper is not to do a comparison for different methods, while we propose a new method from another point of view. With space limitation, we think it’s better to maintain the status quo.
(5) At line 130, I would say “single band” rather than “single frequency”, that might be misleading.
Thank you so much for your suggestions. We keep cautious attitude about the description. You are right, “band” is better than the “frequency”, as we know the GPR frequency are always referred to “frequency band”. But if we use “single frequency profile”, I think most readers will understand the profile is acquired with a single frequency antenna. Referring to published papers, maybe “single frequency” is a better choice in the context of data fusion. Of course, maybe we will change our idea in the future.
