GEO-RADAR LNAPL MODEL BY ARTIFICIAL INTELLIGENT

Abstract

Ground Penetrating Radar (GPR) as non-destructive measurement of full-wave electromagnetic(EM) backscatter completely relies on dielectric permittivity. Interpreting GPR reflection configuration is a complex qualitative with positioning and depth determination would be misleading due to severe polarization and velocity mismatch in travelling-wave. As a result of these studies, a GPR signal segmentation algorithm model was developed to map and identify light non-aqueous liquid (LNAPL) contaminated in laterite soil utilizing dielectric permittivity prediction. Simultaneous registration of a Global Positioning System (GPS) signal was performed while acquiring georeferenced GPR data sets to pinpoint the appropriate location of the soil layers. In this way, georeferenced GPR dispersion was assessed and dielectric permittivity was retrieved by velocity extraction. Empirical model relationship was established by the higher-order regression. Calibration function used verification measurement with root mean square error (RMSE) and calibrated Performance Network Analyzer (PNA). Segmentation and classification using Support Vector Machine (SVM) classifier as Artificial Intelligence (AI) was executed using predicted dielectric permittivity to construct the GPR automated recognition model. The model was compared with actual data and logistic regression classification. The result shows both classification techniques have provided good quality with root mean square errors (RMSE), which were 0.1391 and 0, respectively. The classification produces correct instances classified above 98% for SVM and 100% for logistic regression.