应用变分模态分解和随机森林特征选择算法的扬声器异常声分类  

Loudspeaker abnormal sound classification using variational modal decomposition and the random forest feature selection algorithm

作　　者：周静雷[1] 周智 崔琳[1] ZHOU Jinglei;ZHOU Zhi;CUI Lin(School of Electronics and Information,Xi’an Polytechnic University,Xi’an 710600,China)

机构地区：[1]西安工程大学电子信息学院,西安710600

出　　处：《振动与冲击》2022年第20期277-283,共7页Journal of Vibration and Shock

基　　金：国家自然科学基金青年项目(61901347);陕西省教育厅科技计划项目(18JK0342)。

摘　　要：为了提高扬声器异常声分类的精度,提出了基于变分模态分解(variational mode decomposition,VMD)和随机森林特征选择算法的扬声器异常声分类方法。首先利用VMD分解采集到的扬声器声响应信号,之后对得到的一系列模态分量提取时域和频域特征;然后利用随机森林特征计算提取特征的重要性,通过递归特征消除算法提取出相关性较强的特征构造出最优特征子集;最后将最优特征子集输入至随机森林分类器中,实现扬声器异常声的分类识别。试验结果表明,该方法可以筛选出规模较小且识别度较高的低维特征数据集,同时具有更好的平均识别准确率,平均识别准确率为98.61%。In order to improve the accuracy of speaker abnormal sound classification,a speaker abnormal sound classification method based on variational mode decomposition(VMD)and the random forest feature selection algorithm was proposed.Firstly,VMD was used to decompose the collected loudspeaker acoustic response signal,and then the time-domain and frequency-domain features of a series of modal components were extracted.Then,the importance of feature extraction was calculated by using random forest features,and the features with strong correlation were extracted by a recursive feature elimination algorithm to construct the optimal feature subset.Finally,the optimal feature subset was input into the random forest classifier to realize the classification and recognition of speaker abnormal sound.Experimental results show that this method can screen out small and high recognition degree of low dimensional feature data sets,and has better average recognition accuracy,with an average recognition accuracy of 98.61%.

关 键 词：扬声器 异常声分类 变分模态分解(VMD) 特征选择 随机森林 

分 类 号：TN912[电子电信—通信与信息系统]