Feature Extraction of Underwater Acoustic Target Signals Using Gammatone Filterbank and Subband Instantaneous Frequency

Abstract

Feature extraction is the key process of underwater acoustic target recognition. Inspired by human auditory perception, several auditory-based features were developed for robust underwater acoustic target recognition. However, most of them only represent the spectral envelope or energy of the signal. Nevertheless, the phase represented by instantaneous frequency (IF) also reflects some characteristics of the target. Furthermore, the IF-based features can be combined with the energy-based features to enhance the recognition performance. In this paper, we propose several features that are extracted from the outputs of the Gammatone filterbank. The combined features are constructed explicitly by concatenating the energy-based and subband IF-based features or implicitly by calculating the energy weighted mean subband IF features. Recognition experiments are conducted on real underwater acoustic target radiated noise signals, and white Gaussian noise is added for simulating different signal-to-noise ratio conditions. Support vector machine is employed as the classifier. The experimental results reveal that the proposed features can enhance the recognition performance compared with the conventional auditory-based features.