Underwater signals' characteristics of power spectrum based on data mining

Abstract

Power spectrum was always utilized for underwater signal recognition. Distance based data mining was applied to feature extraction of power spectrum. The distances within and between classes were first calculated, then, the weight for each frequency bin was get according to its discrimination, finally, the weighted power spectrum was used to feature extraction and classification. All kinds of signals were generated with underwater acoustic propagation model through wave guide. Experiments and tests were conducted and performance of the method was verified.