Target confirmation architecture for a buried object scanning sonar

Efficient mine clearing operations are essential for maintaining sea lines of communication and for the timely dispatch of military and economic supplies to conflicted areas. To locate stealthy buried mines, a future naval system of systems is under development that incorporates high-resolution acoustic and electromagnetic sensors. This paper describes an evolving target confirmation architecture for this program's buried object scanning sonar that utilizes image and signal classification strategies. Feature extraction from the 3-D sediment volume imagery is described. Image classification using a joint Gaussian Bayesian classifier is demonstrated with synthetic 2-D image classification experiments that employ image clustering and ellipse feature extraction methods. The signal classifier is demonstrated with data collected from mine-like and clutter objects buried in sand. These tests utilize data from different run orientations and transmit angles for training, cross validation and testing, achieved 5-class classification levels of 94% and 2-Class ROC curve knee values of Pcc=96% and Pfc=4%, and thus illustrate the buried mine-hunting potential for time-frequency based signal classification.