Robust target feature extraction based on modified cochlear filter analysis model

The performance of underwater target recognition systems depends on the consistency and adaptation of target feature in complex conditions during the training and testing stages. In this paper, we investigated the target feature extraction problems in a complex underwater environment and proposed a novel approach based on modified cochlear filter analysis model of the human auditory system. The frequency responses and distributions of the modified cochlear filter bank are similar to that of the basilar membrane in the cochlea. The fast forward and inverse transforms of modified cochlear filter bank are also presented for discrete-time signals to save the computational load. The experimental results using real ship radiated noise data verified that the modified cochlear filter analysis model is robust in background noise.