Feature Vector for Underground Object Detection using B-scan Images from GprMax

One common technology for underground object detection is Ground Penetrating Radar (GPR). For landmine detection, it is vital to have a fast and accurate method. This paper uses synthetic data from GprMax program and proposes a 3-step method to locate and discriminate underground objects: 1) Pre-processing using n-rows average 2) Image scaling and 3) converting Region of Interest (ROI) to a feature vector. Proposed method has been tested using 5 methods; 2 classification algorithms; and 3 different image scales. The detection accuracy and runtime performances have been reported according to classifiers. Proposed method has a good potential with its runtime performance and small representation capacity. Although, it has slightly lower performance for K-Nearest Neighbors (KNN) compared to Histograms of Oriented Gradients (HOG), proposed method increases overall performance comparably for Support Vector Machines (SVM) from 67.6% to 85.5%.