Simultaneous feature and HMM Model learning for landmine detection using Ground Penetrating Radar

Hidden Markov Models (HMMs) have been widely used in landmine detection with Ground Penetrating Radar (GPR) data; however, to the best of our knowledge, there are no other studies that investigated the simultaneous learning of the features and the HMM parameters. In this paper, we present a novel method based on Gibbs sampling that both learns a feature extraction model as well as an HMM model. The new system allows for the training of new features when the sensor systems are different. Experiments show that our algorithm is more robust to initialization and can find better solutions.