Model based detection and identification of land-mine signatures in GPR-data

During the past couple of years, several GPR data sets have been acquired at the TNO-FEL test-site ??Waaisdorp??. Sinultaneously a detection algorithm, based on autoregressive mandelling has been developed and implemented. This algorithm first determines a measure of the information content of data from known mine Ideations, by means of minimum description length (MDL) analysis. The result determines the number of poles needed for building autoregressive models of land-mines. These maldels are then used to determine the statistical distance with respect to GPR measurements. Where this stochastic distance is minimal we declare a detection. We briefly introduce the theoretical background behind the detection algorithm and indicate its use and application. The algorithm is subsequently applied to various datasets representing different environmental scenarios and land-mines. The detection results are presented and influence of radar specific parameters and possible environmental factors am: also discussed.