基于EMD和2（1/2）维谱的直升机声信号特征提取

利用经验模态分解（Empirical Mode Decomposition,EMD）的方法将直升机声信号进行分解,得到一系列本征模态分量（Instrinsic Mode Function,IMF）。计算实际直升机声信号及由其分解得到的每个IMF分量的四阶累积量对角切片谱,并由此得到实际信号及每个IMF分量的四阶累积量对角切片谱的幅度绝对值之和E。计算每一个IMF的E值与实际信号E值的比值构成直升机声信号特征矢量。采用神经网络分类器,对两种不同机型的直升机声信号进行分类和识别。仿真实验验证了该方法是可行的、有效的,分类识别取得了较好的效果。

Feature Extraction of Helicopter Acoustic Signals Based on EMD and 21/2 Dimension Spectrum

Empirical mode decomposition（EMD） has been used to decompose the helicopter acoustic signal,and a series of instinstic mode function（IMF） have been got.Compute 4-th order cumulant diagonal slice（FOCS） spectrum of the actual helicopter acoustic signal and its IMFs separately,and then each sum of absolute ampnitude of FOCSS can be calculated.Calculate the ratio of E of each IMF to that of the actual helicopter acoustic signal and chose it as a member of eigenvector.Two different kinds of helicopter acoustic signal are classified and recognized by BP neural network.The simulation results shows that the proposed method is viable and effective,the classification and recognition results is satisfactory.