The theory and model of auditory peripheral has made great progress

and has been widely used in speech signal processing.The text combined Gammatone filter and Meddis model to simulate processing mechanism of cochlear

and modified parameters of Meddis model according to the characteristics of underwater acoustic signals.Based on the Gammatone-Meddis auditory peripheral model

an approach to extract features from underwater target signal was presented.Then a twenty-three dimension feature vectors were obtained.With the feature vectors

the training and recognition processes was performed.The real sea experiments show that the features can classify effectively and the recognition ratio may arrive 94% above

that the features are robust and the capability of recognition does not decline when convolution noises are added to original signals.Finally

the experiments prove that both nonlinear selecting frequency by basement membrance and inner hair cells can restrain noise.