Communication signal analysis with fusion classification of captive bottlenose dolphins  

作　　者：ZHANG Xiaowei ZHANG Chunhua XUE Shanhua YIN Li 

机构地区：[1]Key Laboratory of Science and Technology on Advanced Underwater Acoustic Signal Processing,Institute of Acoustics,Chinese Academy of Sciences,Beijing 100190 [2]University of Chinese Academy of Sciences,Beijing 100049

出　　处：《Chinese Journal of Acoustics》2023年第2期139-152,共14页声学学报（英文版）

基　　金：supported by the"Sea Program"Project of Qingdao Pilot National Laboratory of Marine Science and Technology(SQ2017WHZZB0701-3-2)。

摘　　要：For improving the classification accuracy of bottlenose dolphin communication signals(whistle)under captive conditions when overlapping echolocation signals(click),a fusion classification method based on machine learning is proposed.The time-frequency distribution features of whistle signals are extracted to train the random forest classifier,and the Mel timefrequency diagram features are used to train the convolution neural network classifier.On this basis,a fusion decision-maker is designed to classify and recognize the aliased whistle signals.The classification and recognition results of the experimental data collected from the sound signals of captive dolphins showed that the fusion classification method has better classification performance.The classification accuracy of the aliased whistle signals is more than 94%,which is better than the time-frequency distribution feature classifier and Mel time-frequency graph feature classifier,and can improve the classification ability of the aliased signals.

关 键 词：SIGNAL CLASSIFIER DISTRIBUTION 

分 类 号：Q958.8[生物学—动物学] TP181[自动化与计算机技术—控制理论与控制工程]