Underwater Noise Classification based on Support Vector Machine

In the face of more and more underwater noise monitor data, there is a need to process the noise automatically. In this paper, support vector machine (SVM) classifiers with different kernel functions are adapted to distinguish five kinds of underwater noises. At the same time, five kinds of features are extracted from frequency domain, time-frequency domain and Mel transform domain, including noise spectrum level, time-frequency spectrum, power spectral density, Mel-scale Frequency Cepstral Coefficients and filter-bank. According to 7225 samples with a duration of 0.5s and sampling frequency of 44.1 kHz, the results show that treating the kernel function as a "black box" is suitable for underwater noise classification. It can obtain appropriate parameters adaptively and improve the performance. Among the five input features, the classifier performs better with the input features of NL and PSD. Under the condition of -10dB, the mean classification accuracies of NL and PSD features are 93.63% and 94.53%, respectively.