Model-based target classification using spatial and temporal features of metal detector response

The paper presents a novel model-based algorithm for classifying buried metallic targets using spatial and temporal response properties of a pulse induction metal detector mounted on a mobile robot for autonomous landmine detection. In the proposed approach, we firstly derive a simplified analytical model for spatial distribution of the primary magnetic field that corresponds to transmitter/receiver coil geometry of a given metal detector. The sensing head model is then coupled to a metallic target analytical dipole model whose parameters are the magnetic polarizability tensor and the target location. Finally, the forward sensor/target model is fitted to sensor data obtained by spatially mapping the suspected target area using a mobile robot. Inverted magnetic polarizability tensors corresponding to sensor data acquired at different time instances (gates) are used for target characterization and classification. The algorithm is experimentally evaluated on a dataset collected from a test site containing surrogate mines (metallic spheres) and clutter targets.