Autonomous interpretation of side scan sonar returns

The automatic interpretation of side scan sonar (SSS) returns is addressed. The proposed system combines signal detection techniques, feature extraction using SSS geometry, and a neural network classification algorithm for real-time detection and classification. Preliminary results were obtained by experimenting with actual SSS data. Applying the automated system to the data, the detection subsystem was able to detect properly the targets of interest as well as other bottom objects. The feature extraction subsystem helped the neural classifier separate targets from extraneous objects. However, in many instances, more resolution was needed to describe small targets accurately so that the neural classifier could classify correctly.<
>