Texture analysis of GPR data as a tool for depicting soil mineralogy

Abstract

For the past decade, texture analysis has become one of the quantitative attributes used extensively in seismic studies for target detection and interpretation of subsurface anomalies, such as moisture content and landmines in inhomogeneous soil. The application of this analysis has been limited to other geophysical techniques, such as Ground Penetrating Radar (GPR). GPR is a non-invasive method based on the propagation of electromagnetic waves to derive a model for the subsurface. In general, interpreting GPR data is primarily qualitative and depends on the expertise of the analysts. The goal of this study is to verify the ability of texture analysis technique to differentiate soil mineralogy. We have developed and tested a Matlab code that derives various texture measures such as energy, homogeneity, contrast, and entropy. Those statistical measures are generated using a gray level co-occurrence matrix (GLCM). The measures supply different information about the data, such as uniformity or complexity; thus, they can produce different features in the GPR data when they are used together. We tested the texture analysis code on synthetic GPR data sets that have been derived for two different models with two different heavy minerals, ilmenite and spodumene embedded inside a host medium. To obtain synthetic data, we used GPRMax2D software which applies the Finite Difference Time-Domain method (FDTD) to simulate various subsurface scenarios. In addition to the synthetic data, real GPR data that had been collected from a prototype laboratory experiment were used. The calculated statistical features and results show that ilmenite has higher entropy, dissipation, and contrast measures than spodumene. On the other hand, spodumene shows higher energy and homogeneity features than ilmenite. Based on the synthetic data results, the combination of the texture-based analysis measures can be used as an enhanced interpretation tool that clearly brings out the distinction between minerals. The texture results computed from ground-truth GPR data show that heavy mineral bodies can be identified due to their high contrast, entropy, correlation, standard deviation, and low energy and homogeneity. Variance measure of texture analysis can helphighlight the edge of the buried samples. Cluster indicator is more effective in visualizing the anomaly than the raw data.