Automatic classification of metallic targets using pattern recognition of GPR reflection: a study in the IAG-USP Test Site, Sao Paulo (Brazil)

In this work, a methodology to automatically classify of metal targets using pattern recognition techniques on GPR reflection data is presented. The methodology consists of designing a multilayer perceptron (MLP) classifier based on features extracted from the targets in the subsoil, and then using it to classify hyperbolas diffraction indicating their position and depth. The classification of reflections allows a high resolution reconstruction of the subsurface with reduced computing time. The system was developed in MATLAB and applied to GPR data obtained at IAG-USP test site, located in the city of Sao Paulo, Brazil, where metallic drums were studied under controlled field conditions. This site contains different targets of variable sizes buried under different depths and it served as a model for the computational experiment. The results indicate that the automatic classification of the metallic targets in the subsoil is efficient, contributing for the reduction of the ambiguities in the geophysical data interpretation, besides having application on the subsoil mapping of utilities.