One-class SVM for landmine detection and discrimination

In this paper, we present landmine detection and discrimination method: one class support vector machine (OSVM) based on RBF kernel using one-dimensional Ground Penetrating Radar (GPR) delivered data. The GPR has been a precious tool for humanitarian demining. It scans the ground and delivers a three-dimensional matrix representing three types of data; Ascan, Bscan and Cscan. The Ascan data represents the response from a reflection signal of a pulse emitted by the GPR at a given position. The normalized Ascan data is the input data of our proposed landmine detection method. One Class SVM has been tested on the MACADAM database which is composed of 11 scenarios of target class (landmines) and 5 scenarios of outliers class (wood stick, Soda Can, pine, stone), each evaluation scenario contains six buried objects in various buried depth which varied between −70 and 100 mm. OSVM based on RBF kernel has been compared to the OSVMs based on Polynomial kernel, Linear kernel and Sigmoid kernel in term of classification accuracy. Obtained experimental results which are 89.24% as AUC and 0.959s as running time prove that one class SVM based on RBF kernel is out performs than the others classifiers in terms of landmine detection and discrimination.