Intelligent identification and classification of ground-penetrating radar datasets for sedimentary characterization

The interpretation of Ground-penetrating Radar (GPR) is commonly performed by different GPR experts, resulting in somehow subjective results, especially in complicated backgrounds like depositional sedimentary environments. In order to improve data interpretation, we propose a new intelligent identification and classification strategy based on texture characteristics of GPR to objectively describe and assess subsurface structures. We exploit a K-means++ clustering method to classify different sedimentary units, testing the methodology on a real GPR dataset acquired on the Piscinas dunes, southwestern Sardinia, Italy. The GPR dataset is fully georeferenced and we used not only amplitude data, but multi-attributes extracted using the Gabor filters. Besides, we also evaluate the applicability and feasibility of the Principal Components Analysis (PCA) dimension reduction algorithm to reduce redundant information in dataset selection. The results show that the proposed algorithm can successfully identify and classify the different typical radar facies of subsurface sedimentary structures with an intelligent, objective and repeatable manner, not only identifying sedimentary layering, but also accurately dividing the subsurface sequence in the different depositional and erosional facies.