Abstract: In view of the feature extraction and choice problem of acoustic target in complex environment, based on the time-frequency characteristics of acoustic signals, an effective method of feature extraction for acoustic signal is presented. Firstly, a method of wavelet decomposition is employed for signal de-noising. Secondly, taking the short-time energy, the short-time average amplitude, the zero crossing rate and the energy of signals' frequency bands as initial features, the low-dimensional feature vectors are constructed by combining the Fisher discriminant criterion. Finally, the features of the testing sample data of two types of acoustic targets are extracted, and the validity of the feature extraction method is verified by using support vector machine and K-nearest neighbor classifier. The experimental results show that the feature extraction method of “time domain + frequency domain + linear discriminant analysis” is simple and effective, and it shows higher recognition rate compared with single feature extraction methods.

Key words: wavelet decomposition, feature extraction, linear discriminant analysis, support vector machine, K-nearest neighbor