Correlation based landmine detection technique.

摘要

Landmine detection using Ground Penetrating Radar (GPR) in a handheld unit is a challenging task. The difficulty is due to the inconsistency of landmine signatures and the nonstationary behavior of background clutter. In this work we examine Correlation based detection algorithm for data collected by GPR in standing mode. Standing mode processing uses the first two sweeps to extract feature templates for subsequent detection. This research proposes techniques to improve the current performance of standing mode feature template based detection. It also provides investigation of the performance of the MACE (minimum average correlation energy) correlation filter for landmine detection. This work also studies the performance of the Correlation based landmine detection technique as the size of the template is varied and the possible improvement by combining results obtained from real and imaginary data template. This work concludes by suggesting a new template matching technique for detection.; The proposed algorithm using interpolation provided 10% reduction of false alarm rate at 95% Pd compared to without interpolation. Results of the investigation suggested that the MACE filter has limitation to discriminate between the landmine and clutter data, and the results of the ROC curves showed its difficulty to generalize to the test data. The new template matching technique suggested provided better detection compared to the original method used to perform template matching, when Pd was below 99.5%.