An EM-IMM based abrupt change detector for landmine detection

In this paper, we propose an expectation maximization (EM) trained interacting multiple model (IMM) abrupt change detector for land mine detection applications. The proposed EM algorithm learns the parameters of the different models in real time without requiring a priori information on either the number of models or the model parameters. Using the real ground penetrating radar (GPR) data, the learning performance of the EM-IMM technique is analyzed and commented upon. Numerical receiver operating characteristics (ROC) analysis and detected images indicate that the proposed EM-IMM based abrupt change detector has a better detection and imaging performance than the conventional Kalman filter for land mine detection applications.