中文摘要：

在能检测到天然地震和矿震的区域，这两类地震的快速识别无论对于区域台网和矿区台网都具有现实意义。这两类震动都是非稳态信号，用传统的Fourier变换不能提取出信号的特征信息，小波包分析方法却能很好提取出信号的特征信息。本文提供了一种基于非参数识别算法，即把信号变换到频域，然后再用奇异值分解作为统计工具，提取出信号的特征信息，作为识别天然地震和矿震的识别因子。以辽宁抚顺2001年1月1日到2003年6月30的18个矿震和16个天然地震，以及北京门头沟2001年1月1日到2002年12月31日的15个矿震和14个天然地震为样本，提取出识别因子。最后，用其它的天然地震和矿震资料检验了识别因子的识别率。

英文摘要：

Signal possessing of non-stationary information are not suited for detection and classification by traditional Fourier methods. An alternate means of analysis needs to be employed so that valuable time-frequency information can be extracted. The Wavelet Packet Transform （WPT） is such a time-frequency analysis tool. It describes an approach to signal classification that may satisfy the need of a non-parametric feature extraction algorithm that best adapts to sets of pre-classified data. There are 18mining shocks and 16 earthquake recorded by three-component seismographs in the seismic network in Liaoning Province from Jan. 1, 2000 to June 30, 2003 and 15 mining shocks and 14 earthquakes recorded by three-component seismographs in the seismic network in Huabei from Jan 1, 2000 to Dec 31, 2002. Using the WPT method, a parsimonious set of features is determined from above mining shocks and earthquake data. Finally, the feature set is tested by signals in the waves of many mining shocks and earthquakes.