The Researches On Underwater Targets Recognition Based On The Auditory Model

Automatic passive underwater target classification becomes more and more important. The key step of classification is the extraction of target features. Many researches on auditory model have been successfully utilized on speech processing recently, while on underwater acoustic signal processing has long been neglected. The reason for applying automatic passive underwater target classification based on auditory model that the feature extraction of auditory model is an important part of underwater acoustic signals processing. This paper concentrates on the research on human auditory system dealing with the sound signal. Also the ship-radiated noise in real situations is studied and proved separately based on the application of wavelet analysis in denoising, feature extraction and targets recognition.The main work and originality of this paper can be summarized as follows:1. An auditory model suitable for underwater acoustic signal processing is built by studying the success of auditory model in dealing with speech signal. The approach of how to extract underwater acoustic signal with auditory model is studied and the auditory features of ship-radiated noise is extracted.2. The application on wavelet transform to denoising is discussed. The main theory and approach to wavelet transforms are studied. The approach by using wavelet to denosing is applied in real situation and the result shows that it can remove background noise from ship-radiated noise effectively.3. Studies on the feature extraction by wavelet packet transform for ship-radiated noise is present. The energy value of each bind is the feature of ship-radiated noise. The theory background is demonstrated and capability of features is discussed. Three kinds of approaches to extract ship-radiated noise feature are applied, and classified by neural network. The results show that the approach by using auditory model is the best one that achieves the good recognition.4. A neural network classification for target recognition is present. The simulations recognition experiment is performed to prove the capability of features classification on real ship-radiated noise.