Fuzzy processing of hydrophone sounds

Abstract

We describe a system which will process and classify dolphin sounds recorded with a hydrophone. A recently proposed method for determination of fuzzy membership functions on the basis of statistical data is extended to the case of continuous data. This method is an application of mass-assignment theory. The resulting fuzzy sets are used in rules which describe the class structure of the data. The rules are evaluated using the evidential rule of support logic, with weighting parameters derived by semantic discrimination analysis. Our approach should be applicable to a range of signal processing and pattern recognition problems. The methods are introduced with reference to a simple tutorial problem, and then developed to illustrate a more realistic application. A detailed presentation of our methodology is supplemented with discussion of implementation and results. In comparison with over ten alternative methods our approach gives favourable results.