基于小波包能量及高阶谱的特征提取方法  

Feature Extraction Method Based on Wavelet Packet Energy and High-Order Spectrum

作　　者：孙洁娣[1,2] 靳世久[1] 

机构地区：[1]天津大学精密测试技术与仪器国家重点实验室,天津300072 [2]燕山大学信息科学与工程学院,秦皇岛066004

出　　处：《天津大学学报》2010年第6期562-566,共5页Journal of Tianjin University(Science and Technology)

基　　金：国家自然科学基金重点资助项目(60534050)

摘　　要：针对广泛存在的油气管道周边安全问题,研究了管道周围地面活动目标产生的震动信号的特性,提出了一种基于小波包能量谱和信号高阶谱分析相结合的特征提取方法来区分不同的活动目标.根据目标产生的地面震动信号是非平稳的特点,采用基于小波包分解能量的方法对信号的各频带进行分解,得到信号在不同频带内的能量分布特性.仅根据能量谱并不能完全区分不同类型信号,通过对信号高阶统计特性的分析,提取出高阶谱特征频率,结合这两种方法提取出的特征作为神经网络的输入向量进行模式识别.通过对实验数据进行分析,单独采用小波包能量特征其平均识别率为88.5%,而采用本文提出的方法平均识别率可以提高到94.6%,验证了文中提出方法的有效性.In view of the prevalent peripheral security problems around oil and gas pipelines,characteristics of seismic signals generated by moving targets on the surrounding ground have been investigated and a feature extraction method based on wavelet packet energy distribution and high-order spectrum analysis has been put forward to distinguish the different targets. As the seismic signals generated by ground targets were not stable,wavelet packet energy method was used to decompose the signals at several independent frequency bands and to draw the energy distribution features in these bands. However,the wavelet packet energy distribution could not distinguish the types of signals very well. Therefore ,high-order statistic features of signals were analyzed and the characteristic frequencies were extracted. The features extracted based on wavelet packet energy distribution and high-order spectrum were input into the neural network as eigenvectors for pattern recognition. Experiment results indicate that the average recognition rate is 88.5% with only wavelet packet energy spectrum based method while it increases to 94.6% with the method proposed ,thus verifying its effectiveness.

关 键 词：特征提取 小波包能量谱 高阶谱分析 地面震动信号 

分 类 号：TP274.2[自动化与计算机技术—检测技术与自动化装置]