Classification of Buried Targets Using Ground Penetrating Radar: Comparison Between Genetic Programming and Neural Networks

The detection and classification of buried targets such as unexploded ordnance (UXO) using ground penetrating radar (GPR) technology involves complex qualitative features and 2-D scattering images. These processes are often performed by human operators and are thus subject to error and bias. Artificial intelligence (AI) technologies, such as neural networks (NN) and fuzzy systems, have been applied to develop autonomous classification algorithms and have shown promising results. Genetic programming (GP), a relatively new AI method, has also been examined for these classification purposes. In this letter, the results of a comparison between the classification performances of NN versus the GP techniques for GPR UXO data are presented. Simulated 2-D scattering patterns from one UXO target and four non-UXO objects are used in this comparison. Different levels of noise and cases of untrained data are also examined. Obtained results show that GP provides better performance than NN methods with increasing problem difficulty. Genetic programming also showed robustness to untrained data as well as an inherent capability of providing global optimal searching, which could minimize efforts on training processes.