Research on band energy extraction and classification of three kinds of fishes sound signals

The exploitation and utilization of marine fishery resources put forward urgent technical requirements for classification and recognition of economic fishes. According to the different acoustic features of vocal fishes, the classification of vocal signals of three kinds of fishes is implemented by supervised machine learning in this paper. Based on wavelet packet decomposition technique, the band energies of the vocal signals of yellow croaker, rice fish and yellow drum are extracted and classified by different classifiers. The results show that the vocal signals of the three kinds of fishes are mainly concentrated in the frequency bands of 300-800 Hz, and the band energies of vocal signals based on wavelet packet decomposition can achieve effective classification for the three kinds of fishes. In different classifiers, linear discriminant classifier and random subspace discriminant classifier have better classification results. This method can provide services for the exploitation and utilization of marine fishery resources.