飞行状态下短翼菊头蝠与鲁氏菊头蝠回声定位声波小波包识别方法  

Recognition method of wavelet packets for echo-locating calls of flying rhinolophus lepidus and rhinolophus rouxi

作　　者：张新娜[1] 王双维[1] 冯江[2] 胡秀丽[1] 刘颖[2] 何小华[1] 施利民[2] 

机构地区：[1]东北师范大学物理学院,长春130024 [2]东北师范大学城市与环境科学学院,长春130024

出　　处：《声学技术》2008年第1期44-48,共5页Technical Acoustics

基　　金：国家自然科学基金(批准号:30370261);教育部新世纪优秀人才支持计划(NCET-04-0309);教育部重点项目(104257);吉林省杰出青年基金(20030114)

摘　　要：为了对蝙蝠回声定位声波进行种类识别,论文基于离散小波包分解的特征提取方法,对飞行状态下短翼菊头蝠与鲁氏菊头蝠的回声定位声波进行三层小波包分解,提取两种菊头蝠在不同频率带内声波信号的能量作为特征参数,并根据U检验结果选取参数作为识别特征向量,进行BP神经网络识别。其中短翼菊头蝠和鲁氏菊头蝠回声定位声波训练样本分别为95个和102个,测试样本分别为44个和68个。对现有测试样本识别率达到100%。结果表明,基于小波包分析和神经网络的分类方法对蝙蝠回声定位声波进行识别是可行的。In order to classify bat＇s echo-location to species level, a feature extraction based on wavelet packet decomposition is used. Three wavelet packets decomposition is applied to the echo-locating calls of flying Rhinolophus Lepidus and Rhinolophus Louxi. The energy values in different frequency bands of sound signal are extracted as characteristic parameters. An eigenvector made up of appropriate parameters according to the U testing result is used to recognize the two bats with Back Propagation Neural Network. In this paper, the numbers of training calls of Rhinolophus Lepidus and Rhinolophus Louxi are 95 and 102 respectively, and the numbers of testing calls are 44 and 68 respectively. The correct classification rate of existing testing calls can get up to 100% for the sample signals. The result shows that the method based on wavelet packet analysis and artifical neural network is feasible to recognize bats.

关 键 词：蝙蝠 小波包 特征提取 神经网络 识别 

分 类 号：O429[理学—声学]