GPR identification of voids inside concrete based on support vector machine (SVM) algorithm

Abstract

In this paper, voids inside concrete is identified and classified from ground penetrating radar (GPR) image in a completely automatic way basing on the support vector machine (SVM) algorithm. The entire process can be characterized into four steps: 1), the original SVM classifier model is built up by training the synthetic GPR data obtained by Finite Difference Time Domain (FDTD) simulation, after data preprocessing, segmentation and feature extraction. 2), the classification accuracy of different kernel functions is compared with cross-validation method and the optimization of penalty factor (c) of SVM and the Coefficient (σ <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> ) of kernel functions are obtained by the grid algorithm and the genetic algorithm (GA). 3), this model is then verified and validated by applying to another set of synthetic GPR data, also obtained by FDTD simulation, to test the effect of classification of voids inside concrete. The result shows a high success rate for classification. 4), this original classifier model is finally applied to real GPR data to identify and classify voids. The result is less than ideal when compared with the application to the synthetic data before this original model is improved. In general, this study shows that SVM exhibits promising performance in GPR identification of the voids inside Concrete. Nevertheless, the recognition of the shape and distribution of voids may need further improvement.