Time-frequency representations for classification of ground penetrating radar echo signal

Abstract

This paper deals with the classification of ground penetrating radar (GPR) echo signal using time-frequency representations. The algorithm fulfills automatically feature extraction and classification of the ground penetrating radar echo signal. We first use auto-ambiguity function as signal representation, which does not use time-frequency smoothing kernel to reduce cross-term. Cross-term is interference for visualization interpretation, but for classification it may be valuable information. Then we use Fisher's discriminant ratio to rank the discriminant information of features, and in combination with using classification error rate of learning vector quantization neural network as evaluation function to select optimal features subset. Experimental results based on simulated and measured GPR data are presented.