EEMD样本熵的扬声器异常音分类方法

为了更准确的对扬声器异常音进行分类,提出一种基于集合经验模态分解(EEMD)与样本熵的扬声器异常音特征提取方法并利用极限梯度提升(XGBoost)算法进行分类。在基频陷波预处理后,对信号进行EEMD,结合相关性分析选取固有模态函数(IMF)分量并计算其样本熵构成特征向量。实验结果表明,针对扬声器异常音分类问题,在小样本的情况下,扬声器声响应信号经基频陷波预处理后,XGBoost算法使用EEMD与样本熵的特征提取方法取得了95. 33%的分类准确率,高于小波包变换和样本熵特征提取方法所取得的准确率,验证了特征提取及分类方法的有效性。

Method of classification for loudspeaker abnormal sound based on EEMD and sample entropy

To classify the abnormal sound of loudspeaker more accurately,an algorithm was proposed in this paper,in which ensemble empirical mode decomposition( EEMD) and sample entropy were used for feature extraction and extreme gradient boosting( XGBoost)was used for classification. After preprocessing of pitch notching,the loudspeaker response signal was decomposed using EEMD. The intrinsic mode function( IMF) components were selected with correlation analysis and their sample entropy values were calculated to structure the feature vectors. Focused on the classification for loudspeaker abnormal sound with small sample condition,the experiment results have shown that,after the preprocessing of pitch notching for loudspeaker response signals,XGBoost algorithm with EEMD and sample entropy feature extraction method can accurately classify abnormal sound of loudspeaker,which is more accurate than XGBoost algorithm using feature vectors of wavelet packets and sample entropy,moreover,it achieves 95. 33% classification accuracy.