小波包样本熵的扬声器异常音特征提取方法

摘要

To classify the loudspeaker abnormal sound more accurately, a feature extraction method is proposed, in which wavelet packet decomposition and sample entropy are used. After preprocessing of pitch notching, the loudspeaker response signal is decomposed using wavelet packet decomposition of three levels. Sample entropy values of reconstructed signals are calculated to structure the feature vectors. In the small sample case, the results of experiment show that the SVM algorithm with wavelet packet decomposition and sample entropy feature extraction method achieves 93.33％ classification accuracy.It's 5％ higher than that of energy mean method, which proves the proposed method.%为了更准确地对扬声器异常音进行分类,给出一种基于小波包分解和样本熵的扬声器异常音特征提取方法.在基频陷波预处理后,对信号进行3层小波包分解,计算重构信号的样本熵以构成特征向量.实验结果表明,在小样本的情况下,SVM算法使用小波包分解和样本熵特征提取,分类准确率为93.33％,比能量均值方法高5％,验证了特征提取方法的有效性.