Research On Feature Extraction Of Underwater Diver Breathing-signal

Underwater diver is an increasingly threat to marine security of most countries in the world, and has caused an unprecedented emphasis on naval and maritime sectors; In recent years, navy of most countries have been strengthening the defense strength for underwater diver, and have put forward many techniques and methods of underwater diver detection and recognition based on active sonar. Underwater diver will radiate breathing-signal when performing under missions, by using passive sonar to receive the breathing-signal and establish an effective feature extraction and classification theory, is the core of underwater diver detection and recognition based on passive sonar. In general, there are large numbers of docked or passing vessels existed around the offshore waters like ports and docks, and the sea conditions are complicated, therefore the ambient noise and ship radiated noise under strong background are the main interference factors for detecting underwater diver by using passive sonar. Diver breathing-signal can be regarded as a kind of speech signal according to its produced mechanism, the auditory perception system of human ear has a good ability to perceive the speech signal, its unique advantage of "discrimination based on voice" can complete the perception and recognition of different target signals.The main content of this paper is applying the human auditory perception characteristics to the feature extraction of underwater diver breathing-signal, and complete the classification and recognition between it and ambient noise and ship radiated noise; based on the existed theory of auditory perception parameters feature extraction in the field of nonlinear processing of speech signal and psycho acoustics, this paper selected Mel Frequency Cepstrum Coefficient and timbre parameters(including spectral energy parameters and spectral structure parameters) as the research object, extracted the above feature of breathing-signal, ambient noise and ship radiated noise based on the Mel filter which simulates the auditory perception system:(1)For Mel Frequency Cepstrum Coefficient(MFCC), the paper analyzed the basic principle of homomorphic processing and derived the expression of MFCC, extracted the MFCC parameters of the above three kinds of signals according to the lake trial data, the extracted MFCC parameters were matched and the MFCC angle and MFCC distance between breathing-signal and the two kind of noise based on the matching results were calculated. The results show that the MFCC parameters of divers breathing and the ship radiated noise or ambient noise are obviously different and it can be distinguished between divers breathing signal and noise, the validity of MFCC algorithm can be proved and it has a practical significance to develop the underwater diver detection sonar and warning system near the offshore waters.(2)For timbre parameters, the expression of the spectral energy parameters and spectral structure parameters is derived and the physical meaning of every parameter is analyzed,the timbre parameters in this paper is being describing based on the frequency domain descriptors of timbre, therefore the timbre features were extracted from the second-order statistics of signal, however, breathing sound signal and the noise do not obey Gauss distribution in the frequency domain, and the second-order statistics will bring serious distributed noise and will leave an serious influence on the extraction effect; high order statistics can suppress the distributed noise and improve the signal to noise ratio, and the separability and recognition accuracy between two kinds of signal can be improved, in order to reduce the amount of calculation,this paper extracted the timbre features on the third-order and the fourth-order cumulant spectrum (1.5 dimension spectrum and 2.5 dimension spectrum), analyzed the differences between three types of signals in the feature domain, and the class separability and recognition rate between breathing-signal and two types of noises were calculated by using class&inner class scatter algorithm and support vector machine algorithm; the results show that the average recognition rate between breathing-signal and ship radiated noise and ambient noise under the very algorithm can respectively reach the maximum of 97.02% and 92.25%, supply a certain application value to underwater diver detection and recognition around the offshore waters.