Analyzing and Classifying Fish Vocalization in the East China Sea Coastal Area

Wenbo Li,Ting Zhang,Yanlin Zhou,Wen Xu

DOI: https://doi.org/10.1109/oceans51537.2024.10682341

2024-01-01

Abstract:This paper addresses the processing and classification of raw audio data collected from the coastal region of the East China Sea over a 24-hour period. The methodology involves preprocessing, feature extraction, feature selection, and classification. The raw signals undergo pre-emphasis and bandpass filtering, followed by segmentation techniques to identify individual fish vocalizations. The analysis of 72,324 segmented fish vocalization samples reveals two distinct types, Type A and Type B, constituting 74.60 % and 25.40%, respectively. Feature extraction encompasses time, frequency, and cepstral domains. The random forest algorithm is employed for forward feature selection to reduce data dimensionality and eliminate redundant information. In the final stage, different classifiers are cascaded to enhance accuracy by mitigating potential misclassifications. The paper presents a focused and effective framework for the processing and classification of raw audio data, specifically targeting fish vocalizations in the coastal region of the East China Sea.