A Method of Fish Classification Based on Wavelet Packet and Bispectrum

• Main Authors: Qiao ZHANG, Feng XU, Tao WEN, Tianze YU
• Format: Article
• Language: English
• Published: IFSA Publishing, S.L. 2014-02-01
• Series: Sensors & Transducers
• Subjects: Fish classification; Feature extraction; Wavelet packet; Bi-spectrum; Support vector machine (SVM).
• Online Access: http://www.sensorsportal.com/HTML/DIGEST/february_2014/Vol_164/P_1862.pdf

The complex structure of fish and multispecies composition complicate the analysis of acoustic data. Consequently, it is difficult to obtain a highly accurate rate of classification by using current approaches. A method of fish classification based on the wavelet packet and bi-spectrum is proposed in this paper. To verify this method, firstly, an ex situ experiment has been performed with three kinds of fish: Crucian carp (Carassius auratus), Yellow-headed catfish (Pelteobagrus fulvidraco) and Bluntnose black bream (Megalobrama amblycephale). The backscattering signals of these fishes are obtained. Secondly, the wavelet packet decomposition of backscattering envelop is done, and the energy of main frequency bands which is reconstructed from each node are calculated. Thirdly, the bi- spectrum of envelop which is constructed using the backscattering of main frequency band in order to filtering the high frequency noise, is extracted as the additional feature. The sub-band energy of wavelet packet and the bi-spectrum are combined as the characteristic indicator to describe the fish feature. Finally, three kinds of fish are successfully classified by the RBF support vector machine classifier. The results reveal that the proposed method has a highly accuracy rate of classification at fish with different shapes.