Image and signal classification for a buried object scanning sonar

As US expeditionary forces address the threat of asymmetric weapons, greater efforts are being made toward development of mine countermeasure technologies. To locate stealthy buried mines, a unique adaptive acoustic system is being developed that combines image and signal classification algorithms using data collected with a broadband, multiaspect sediment volume imaging sonar. Initial testing of a joint Gaussian Bayesian image classification method is described, for which scalar and statistical features calculated from a selected image region are combined to optimize correct classification probabilities and minimize false alarm rates. Potential targets are identified in the acoustic image and passed to biomimetic signal classification algorithms that map time-frequency patterns into object class declarations. Two-category classification results are reported for a buried mine shape and clutter objects.