希尔伯特时频谱特征提取方法  

Hilbert time-frequency spectrum feature extraction method

作　　者：韦娟[1] 顾兴权 宁方立[2] WEI Juan;GU Xingquan;NING Fangli(School of Communication Engineering,Xidian University,Xi'an 710071,China;School of Mechanical Engineering,Northwestern Polytechnical University,Xi'an 710072,China)

机构地区：[1]西安电子科技大学通信工程学院,陕西西安710071 [2]西北工业大学机电学院,陕西西安710072

出　　处：《华中科技大学学报（自然科学版）》2021年第1期50-54,共5页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(51675425);陕西省重点研发计划资助项目(2018GY-181,2020ZDLGY06-09)。

摘　　要：为了优化高频率分辨率的希尔伯特时频谱的表达效果,提出一种基于卷积运算思想的希尔伯特时频谱平稳化和自适应增强方法.首先,由希尔伯特黄变换得到高频率分辨率的时频谱和边际谱,通过设定时域和频域平稳因子及权重得到相应的核矩阵,并依照卷积运算的过程对时频谱进行平稳化;然后,将平稳化之后的边际谱值作为时频谱中相应瞬时频率处谱线的增强因子,对时频谱进行自适应增强;最后,用UrbanSound8K数据集提取时频谱图作为特征,利用深度卷积神经网络进行验证实验.仿真结果表明:本方法能够有效提高希尔伯特时频谱的总体表达效果,自适应地改变时频谱的表现形式.In order to optimize the expression effect of Hilbert time-frequency spectrum with high frequency resolution,a method of Hilbert time-frequency spectrum smoothing and adaptive enhancement based on the idea of convolution operation was proposed.First,the Hilbert-Huang transform was used to obtain the time-frequency spectra and marginal spectra with high frequency resolution.By setting the time-domain and frequency-domain smoothing factors and weights,the corresponding kernel matrix was obtained,and the time-frequency spectra was smoothed according to the process of convolution.Then,the marginal spectrum value after smoothing was used as the enhancement factor of the spectral line at the corresponding instantaneous frequency in the timefrequency spectrum,and the time-frequency spectrum was adaptively enhanced.Finally,the UrbanSound8 K dataset was used to extract the time-frequency spectrum as a feature,and the deep convolutional neural network was used to perform the verification experiment.The simulation results show that smoothing method can effectively improve the overall expression effect of Hilbert time-frequency spectrum,and can adaptively change the appearance of time-frequency spectrum.

关 键 词：时频分析 特征提取 希尔伯特黄变换(HHT) 希尔伯特时频谱 平稳化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]