Skip to Main Content
In this letter, we propose an interacting multiple-model (IMM)-based abrupt change detector for ground-penetrating radar (GPR) applications. Ground clutter varies with surface roughness, soil nature, as well as depth of the soil layer, necessitating a multiple-model approach. The IMM is first trained for a chosen number of models and then used to characterize the GPR data. The IMM predictor segments the entire GPR data into regions of identical models and then identifies targets by detecting abrupt changes in model parameters. The number of models is determined using the minimum prediction error criterion. The prediction performance of the IMM predictor is theoretically analyzed, and its detection performance is also evaluated through an receiver operating characteristics analysis to illustrate the improved performance of the proposed detector.