圈养瓶鼻海豚通讯信号分析及融合分类方法  

Communication signal analysis with fusion classification of captive bottlenose dolphins

作　　者：张晓伟 张春华[1,2] 薛山花 尹力[1] ZHANG Xiaowei;ZHANG Chunhua;XUE Shanhua;YIN Li(Key Laboratory of Science and Technology on Advanced Underwater Acoustic Signal Processing,Institute of Acoustics,Chinese Academy of Sciences;University of Chinese Academy of Sciences)

机构地区：[1]中国科学院声学研究所先进水下信息技术重点实验室,北京100019 [2]中国科学院大学,北京100049

出　　处：《声学学报》2022年第5期643-651,共9页Acta Acustica

基　　金：青岛海洋科学与技术试点国家实验室“问海计划”项目(SQ2017WHZZB0701-3-2)资助。

摘　　要：针对圈养条件下瓶鼻海豚通讯信号(whistle)分类时混叠大量回声定位信号(click)导致分类正确率降低的问题,提出了一种基于机器学习的融合分类方法。分别提取whistle信号的时频分布特征训练随机森林分类器,梅尔时频图特征训练卷积神经网络分类器,在此基础上设计融合判决器对混叠whistle信号进行分类识别。对圈养海豚声信号采集实验数据的分类识别结果表明,融合分类方法具有更好的分类性能,对混叠whistle信号分类正确率大于94%,优于时频分布特征分类器和梅尔时频图特征分类器,能够提高混叠信号的分类能力。Aiming at the problem that the classification accuracy of bottlenose dolphin communication signals(whistle)is reduced due to the mixing of a large number of 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 time-frequency 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 show 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 timefrequency graph feature classifier,and can improve the classification ability of the aliased signals.

关 键 词：时频分布 通讯信号 神经网络分类器 时频图 机器学习 分类识别 分类正确率 瓶鼻海豚 

分 类 号：S865.33[农业科学—野生动物驯养] TN912.3[农业科学—畜牧兽医]