基于经验模态分解的目标特征提取与选择

经验模态分解(EMD)是一种新的非平稳时变信号处理方法,可以自适应地将信号的局部特征逐级分解出来。提出了基于EMD的舰船噪声特征提取与选择方法,将本征模态函数(IM F)分量及其瞬时频率作为特征,并选择其判别熵作为特征向量的可分性度量。数值仿真和实际噪声数据处理的结果表明IM F分量和频率可以充分体现目标的特征,具有良好的类别可分性。

Improving Feature Extraction of Ship-Radiated Target Signals with Empirical Mode Decomposition (EMD) and Hilbert Spectrum

Purpose.Ref.2 pointed out that wavelet transform can not self-adaptively perform feature extraction of non-stationary and time-varying signals.The EMD and Hilbert spectrum method proposed by Huang~([3]) can self-adaptively deal with such signals.We are interested in Ref.3's method because we have long engaged in feature extraction of non-stationary and time varying signals radiated from target ships.In the full paper,we explain in detail how to use EMD and Hilbert spectrum;in this abstract we just list the three topics of explanation:(A) brief introduction to EMD and Hilbert spectrum,in which IMFs(Intrinsic Mode Functions) are explained;(B) feature extraction and analysis of target signals based on EMD and Hilbert spectrum;under this topic,esq.(7) through(10) in the full paper are derived;(C) extraction of three types of real signals of target ships;the results are summarized in Table 1 of the full paper.The results in Table 2 indicate that IMFs are effective feature vectors for classification.