摘要
研究了基于关联积分的广义维数谱的定量计算方法,提出了声目标信号的多重分形特征,并对其特征即广义维数谱的有效性进行了分析;同时利用小波变换分析既能反映信号在变换域特性又保留其时域信息的特点,提出基于小波变换的子空间能量及主要能量集中子空间时域信息的特征提取方法,并通过模糊神经网络识别系统对声目标信号的广义维数谱、子空间能量及时域信息的组合特征进行了验证.
The computational method of multifractal dimensions based on related integral is studied. Multifractal dimensions are proposed to extract features of acoustic signal and they are proved to be effective. As the characteristics in the subspaces can be revealed and the information in the time domain can be kept by the wavelet, power distribution of subspaces and time domain characteristics in the power dominating subspace are proposed to extract other features of the acoustic signal. These features are proved to be effective by identification experiment based on fuzzy neural network.
出处
《自动化学报》
EI
CSCD
北大核心
2004年第5期742-746,共5页
Acta Automatica Sinica
基金
国防科技预研项目(420010901.1)资助~~
关键词
多重分形
小波变换
广义维数谱
特征提取
Calculations
Feature extraction
Fractals
Fuzzy sets
Neural networks
Object recognition
Time domain analysis
Wavelet transforms