天然地震与人工爆破波形信号HHT特征提取和SVM识别研究

A study on seismic signal HHT features extraction and SVM recognition of earthquake and explosion

天然地震和人工爆破信号属于非线性非平稳信号,而传统信号分析方法是针对线性系统平稳信号的,本文采用希尔伯特—黄变换(Hilbert-Huang Transform,简称HHT)试图提取可明确区分天然地震和人工爆破事件的波形特征.通过经验模态分解(Empirical Mode Decomposition,简称EMD)把原信号分解为10个左右的本征模态函数(intrinsic mode function,简称IMF),并取前3个IMF分别提取最大幅值对应的周期(T_(Amax))和倒谱平均值(C_(ave))作为信号特征构建特征样本集,该样本集采用严格的模式识别样本划分法进行样本划分,用支持向量机(Support VectorMachine,简称SVM)进行分类识别,识别率介于75%～94%之间.结果表明模态分量的最大幅值对应的周期(T_(Amax))和倒谱平均值(C_(ave))可以作为识别天然地震和人工爆破的有效特征.

Earthquake and explosion are nonlinear signals originating from non-stationary process. However, traditional signal analysis techniques are only suitable to linear and stationary signals. This paper describeds an attempt to extract discriminative seismic wave features from earthquake and explosion recorded signals based upon Hilbert-Huang Transform （HHT）. By Empirical Mode Decomposition （EMD） original complex seismic signals are reduced to about ten intrinsic mode function components （IMF）. In the first three IMF, extract their respective maximum amplituders period （ZAmax）and average cepstrum values（Cave） as signalsr features. The features data matrix of 1583 lines and 6 rows extracted from these signals by the above mentioned extraction method. This sample was partitioned by strict mode recognition samples, and was recognized by Support Vector Classifier （SVC）. About the model which was got from part of the collection of the samples, the prediction capability is very effective, correct recognition rate reached about 75%-94%. This result shows that maximum amplitude＇s period （TAmax）and average cepstrum values （Cave） of EMD components of original seismic wave are effective features for discriminating earthquake and explosion.