Automated approach to classification of mine-like objects in sidescan sonar using highlight and shadow information

The majority of existing automatic mine detection algorithms which have been developed are robust at detecting mine-like objects (MLOs) at the expense of detecting many false alarms. These objects must later be classified as mine or not-mine. The authors present a model based technique using Dempster-Shafer information theory to extend the standard mine/not-mine classification procedure to provide both shape and size information on the object. A sonar simulator is used to produce synthetic realisations of mine-like object shadow regions which are compared to those of the unknown object using the Hausdorff distance. This measurement is fused with other available information from the object's shadow and highlight regions to produce a membership function for each of the considered object classes. Dempster-Shafer information theory is used to classify the objects using both mono-view and multi-view analysis. In both cases, results are presented on real data.