A time frequency domain feature extraction algorithm for landmine identification from GPR data

Ground-penetrating radar (GPR) is a promising technique for demining procedures since it is able to detect both plastic and metal cased antipersonnel landmines. Yet, landmine identification using GPR is a challenging task since other buried reflectors such as stones or metallic debris can be detected. In this paper, a target discrimination approach is analysed and tested experimentally. It is based on relevant features extracted from 1D, GPR signals in the time frequency domain using the Wigner-Ville distribution. Firstly, a filtering algorithm is applied to remove unwanted reflections from the background (e.g., soil surface reflection). Secondly, the preprocessed signal is transformed into a time frequency image using the Wigner-Ville distribution. Finally, relevant features are extracted based on the singular value decomposition of the time frequency distribution. The algorithm is tested on radar data collected using two different hand-held systems: (i) an impulse GPR-based dual-sensor system and (ii) a stepped-frequency continuous-wave GPR. Data were acquired over different types of soil and for different landmines and objects. Results are compared to features extracted using the wavelet transform and the Wilk's lambda value is used as a criterion for optimal discrimination. Promising results are obtained, which show that time frequency features from Wigner-Ville distribution could be used for differentiating between landmines and false alarms and could contain more valuable information than the features extracted using wavelet transform.