Autonomous Sonar Classification Using Expert Systems

An expert system can process active sonar returns, perform geometric analysis and autonomously classify detected underwater objects. Autonomous classification of objects is an essential requirement for independent operation by autonomous underwater vehicles (AUVs). Most AUVs are only capable of rudimentary sensor analysis, since standard approaches to evaluation and classification of sonar data require excessive signal processing and computational power to be practical. This paper describes how to develop an autonomous sonar classification expert system for a working AUV. A fundamental approach is presented for applying geometric reasoning and expert system heuristics to sonar classification. Preliminary sonar processing is performed using parametric regression line fitting. A polyhedron-building algorithm correlates the parametric regression line segments into geometric objects. After quantifying geometric object attributes, objects are classified using rule-based evaluation of quantitative and qualitative attributes combined with sonar classification heuristics. A summary of expert systems describes their salient features pertinent to autonomous sonar classification systems. The expert system paradigm, knowledge representation, reasoning using facts and rules, rule sets, control of execution flow and expert system development are outlined. Expert system self-diagnosis and self-correction are also discussed. Implementation was performed using the "C" Language Integrated Production System (CLIPS) expert system shell. Real-time graphic simulation and scientific visualization are employed to evaluate results. Experimental sonar classification results are presented using actual mission data from the Naval Postgraduate School (NPS) AUV. Successful classifications of walls and a mine-like object are demonstrated.