A Spatial Inference Approach for Landmine Detection Using Forward-Looking GPR

We propose to detect landmines and unexploded ordnance in forward-looking ground-penetrating radar imagery by applying a spatial multiple hypothesis testing method. Homogeneous regions in an investigation area are identified based on spatial proximity and similarity of the observed decision statistics in order to discriminate targets against clutter. The proposed method is designed to control the proportion of false alarms among all pixels declared to be associated with a target. The detection performance of the proposed method is evaluated using numerical data of shallow-buried targets and compared to existing approaches.