Markov random fields for target classification in low frequency sonar

Low frequency active sonar, when operated in shallow water can suffer from a large number of false "clutter-like" returns. We have used a Markov random field (MRF) approach in order to reduce the number of such false detections by distinguishing between target-like contacts and background in a sonar environment. The model is shown to be based on a sound physical and probabilistic foundation which leaves only a few user defined parameters. The algorithm is used to process real data and results are presented for the variation of a number of clutter objects with model parameters. The results of these tests are presented and we show that the method, for the data investigated, reduces the number of false alarms significantly without loss of target detectability.