Classification of Underwater Acoustic Target Using Auditory Spectrum Feature and SVDD Ensemble

Underwater Acoustic Target Classification (UATC) is an important field in underwater acoustic signal processing. Developing a robust, efficient and high-performance classifier is challenging due to the complex ocean environment, to the great variety of underwater acoustic target classes, and to the ambiguity that arises because of similar classes. Furthermore, in the case of only one class target data is available, the classification model has to be learned from one class samples. In this paper, we proposed a hybrid method combining auditory spectrum and support vector data description ensemble for One Class Underwater Acoustic Target Classification (OCUTAC). The proposed new method consists of 3 stages: (1) Feature extraction method which is based on auditory spectrum for underwater acoustic signal processing. (2) Auditory Spectrum Features extracted from different training datasets which obtained via bagging method are used to construct different individual SVDD classifiers. (3) Individual SVDD classifiers are combined into a single module using a majority voting approach to make a final decision. Experimental results with real data show that the proposed hybrid method is effective for One Class Underwater Acoustic Target Classification.