应用变分模态分解及能量熵的扬声器异常声分类

为更准确地实现扬声器异常声分类以及促进其分类的自动化,提出一种基于变分模态分解(Variational Mode Decomposition,VMD)能量熵和遗传算法优化的支持向量机(Genetic Algorithm-Support Vector Machines,GA-SVM)的扬声器异常声分类方法。首先对测得的扬声器单元声响应信号进行VMD,然后提取每个变分模态函数(Variational Mode Function,VMF)的能量熵并进行统计分析,最后利用GA-SVM进行异常声判断。实验结果表明,与VMD时频熵、经验模态分解(Empirical Mode Decomposition,EMD)能量熵、EMD时频熵这3种特征提取方法相比,VMD能量熵能更准确地表征扬声器单元异常声特征,具有更高的平均识别率,其平均识别率为96.3%,较以上3种方法分别提高了18.3%,24.0%,54.3%。 

Loudspeaker abnormal sound classification using variational mode decomposition and energy entropy

In order to realize the loudspeaker abnormal sound classification more accurately and promote the automation of classification,a classification method of loudspeaker abnormal sound based on Variational Mode Decomposition(VMD) energy entropy and Support Vector Machine optimized by Genetic Algorithm(GA-SVM) is proposed.First,VMD is used to decompose sound response signals of loudspeaker units,then the energy entropy of each Variational Mode Function(VMF) is extracted and analyzed statistically,and finally the judgment of loudspeaker abnormal sound is performed by GA-SVM.The experimental results show that,compared with VMD time-frequency entropy,Empirical Mode Decomposition(EMD) energy entropy and EMD time-frequency entropy,VMD energy entropy can more accurately characterize features of loudspeaker abnormal sound and has higher average recognition rate,which reaches 96.3%,compared with the above three methods,18.3%,24.0% and 54.3% are increased respectively.