Rubost Feature for Underwater Targets Recognition Using Power-Normalized Cepstral Coefficients

Due to the instability of the real underwater environment, auditory processing methods are proved to be effective measures for the acoustic target recognition. As MFCC and GFCC might show inevitable failure with the noise disturbance, in this paper we apply a recent novel feature extraction called Power Normalized Cepstral Coefficients (PNCC) on the underwater targets recognition. Additionally, we propose an enhanced PNCC by adding the Subband mean IF feature. Subsequent experiments show that the PNCC has stronger performance in terms of anti-noise interference than conventional auditory feature and the modified version can improve the recognition accuracy than original one.