Passive acoustic detection of diver based on SVM

In terms of divers equipped with open-circuit respirator, this paper studies the radiated mechanism of the divers' breathing signals. A SVM(Support Vector Machine) based approach to identify their radiated signal is proposed. In this method, the received signal is firstly processed using STFT(short-time Fourier transform). Then, the energy characteristics for sub-bands associated with the remarkable frequency is extracted to reduce the effect of random noise. Lastly, SVM-based method is applied to train many samples to obtain classifier related to classifying the diver. Based on the pool experiment, the comparison in accurate rate of recognition between the proposed SVM method and NMF (Normalized multi-band matched filter) based on setting threshold value in frequency domain is conducted. In addition, the problem of low efficiency of discriminating threshold value in detection in the case of irregular breath of diver is discussed. It is noted that passive diver recognition mainly depends on the divers' breathing signals and the SVM method has higher accuracy than the NMF according to the same set of data.