融合MFCC和IMFCC特征的电厂设备声音识别算法  

Power plant equipment sound recognition algorithm fusing MFCC and IMFCC features

作　　者：翟永杰[1] 彭雅妮 杨旭 胡东阳 王新颖 ZHAI Yongjie;PENG Yani;YANG Xu;HU Dongyang;WANG Xinying(Department of Automation,North China Electric Power University,Baoding 071003,China;Department of Computer,North China Electric Power University,Baoding 071003,China)

机构地区：[1]华北电力大学自动化系,河北保定071003 [2]华北电力大学计算机系,河北保定071003

出　　处：《现代电子技术》2022年第8期6-12,共7页Modern Electronics Technique

基　　金：国家自然科学基金资助项目(61773160)。

摘　　要：为实现非接触式的电厂设备状态监测,文中提出一种将梅尔频率倒谱系数(MFCC)和翻转梅尔频率倒谱系数(IMFCC)两种特征中的部分特征联合作为特征参数,利用多分类支持向量机(SVM)进行参数分类的识别方法。该方法首先对数据集中的声音信号进行预处理操作;然后通过两种特征提取方法生成特征参数,计算生成两种特征每一维的类内均值和类间方差;最后根据阈值选择特征,对选择的两种特征进行线性叠加,从而得到融合特征。实验结果表明,在ESC-50部分数据集和电厂采集数据集上,相对于另外两种特征,融合特征维数更少,识别率更高,并且在训练样本较少的情况下能达到更好的分类效果。In order to achieve non-contact condition monitoring of power plant equipments,a recognition method using multiclassification support vector machine(SVM) is proposed for parameter classification,which combines some features of Mel frequency cepstrum coefficient(MFCC)and inverted Mel frequency cepstrum coefficient(IMFCC)as feature parameters. In this method,the sound signals in the data set are preprocessed,two feature extraction methods are used to generate feature parameters,and the intra-class mean value and inter-class variance of each dimension of the two features are generated by calculation. The feature is selected according to the threshold,and the two selected features are linearly superimposed to obtain the fusion feature. The experimental results show that in comparison with the other two features,the fusion features on the ESC-50 partial data set and the data set collected in power plant have fewer dimensions and higher recognition rates,and can achieve better classification effect in the case of less training samples.

关 键 词：特征融合 声音识别 参数分类 信号预处理 非接触式监测 特征参数生成 状态监测 数据采集 

分 类 号：TN931[电子电信—信号与信息处理]