Metal shape classification of buried object using multilayer perceptron neural network in GPR data

Abstract

Ground penetrating radar (GPR) is a non-destructive tool for subsurface investigation becomes an emerging trend in many applications especially in detecting infrastructure utilities. However, interpreting of GPR data in terms of target’s geometry is remained a challenging task due to the existence intra/inter-class variations and subtle changes of hyperbolic signature pattern in GPR data. To addressed this issue, this paper proposes to classify the basic shape (i.e. cubic, cylindrical and disc) of buried object using multilayer perceptron neural network. The GPR raw data firstly is pre-processed based on A-scan and B- scan GPR data. Then, the combined statistical features with gray level co-occurrence matrix (GLCM) and Hu moment invariants are extracted from A-scan and B-scan. The extracted features are then fed as input to multilayer perceptron to classify the shapes (cubic, cylindrical and disc) of buried object. Based on the experiment conducted, the highest recognition rate has been achieved is about 83.3%. Thus, the proposed method of using combined statistical with GLCM and Hu moment invariants features to classify shape of buried object using multilayer perceptron is promising.