用11/2维谱结合小波包能量提取地震动信号特征

Feature Extraction Method of Seismic Signals Based on 11/2 Dimensional Spectrum and Wavelet Packet Energy

针对车辆(轮式车、履带式车)引起的地震动信号中,具有非平稳、非高斯性特征相互重叠的实际情况,研究了地面活动目标产生的地震动信号特性;从理论上说明了11/2维谱可消除车辆引起的地震动信号中的高斯白噪声或有色噪声,在将11/2维谱分析和小波包能量谱相结合的基础上,提出一种特征提取方法,以便区分不同的车辆目标。在时频域构建以11/2维谱和小波包能量谱作为地震动信号的联合特征向量,建立以训练误差为目标的BP神经网络模式分类器;然后对两类车辆信号进行识别。地震动信号的车辆实测数据表明,该方法能够准确和有效地识别车辆引起的地震动信号。

Aiming at the problem of the overlap of non- stationary and non- Gaussian characteristics in seismic signalsgenerated by vehicles （wheeled vehicles and tracked vehicles）, the features of seismic signals generated by ground movingtargets are studied. It is elaborated theoretically that the 112 dimensional spectrum can eliminate Gaussian white noise orcolored noise in vehicle induced seismic signals. A feature extraction method combining the 112 dimensional spectrum analysiswith wavelet packet energy spectrum is proposed to distinguish different vehicle targets. At first, the joint eigenvectors of 11/2dimensional spectrum and wavelet packet energy spectrum are constructed in time-frequency domains as the seismic signals.Then, the BP neural network pattern classifier with error training as a target is established to identify the two types of vehiclesignals. The results of field experiments show that this method can identify the seismic signals effectively and accurately.