Robust underwater target recognition based on gammatone frequency discrete wavelet coefficients (GFDWC)

In this paper, aiming at the complexity of underwater target-radiated noise and the limitations of study samples, we present a novel recognition feature approach based on GFDWC, which imitates the hearing perception mechanism of humans. Compared with Mel-frequency cepstral coefficients (MFCC) used extensively at present, this method improves the performance of underwater target recognition system in the following two aspects: (1) the triangular filters were improved based on gammatone function, which approximates well the filter response of basilar membrane; (2) taking localization characteristics into account, we replaced the discrete cosine transform (DCT) with discrete wavelet transform (DWT). The results of the recognition experiments of real underwater target-radiated noise and their analysis show preliminarily that the proposed method has good robustness and gives high recognition rates.