核爆地震波的时频特征

Feature extraction on time-frequency plane for seismic pattern recognition of nuclear explosion

对地下核爆炸产生的地震波信号进行联合时频分析 ,旨在研究核爆炸与天然地震信号在时频平面上的差异 ,为核监测提供鉴别判据。除将传统的时域、频域特征统一并推广为时频平面上的矩特征和复杂度特征之外 ,还提出了利用互模糊函数图像 ,时频平面上的时频面积比和累积分布等作为特征 ,对地震波进行更细致的分析。此外 ,将时频表示直接进行奇异值分解 ,用得到的少量奇异值作为一种正交变换特征 ,以减少特征空间的维数。

In this paper,based on time frequency analysis of the seismic signal,feature extraction on time frequency plane is investigated for seismic pattern recognition of nuclear explosion.Firstly,traditional features in time/frequency domain,such as complexity,spectral ratio,and third moment of frequency,are unified and generalized to the features of time frequency domain,called as time frequency moments and complexities. Theoretical analysis and experimental results show that cross ambiguity function is fitter for monitoring and locating of nuclear explosion than correlation function.Moreover,time frequency area ratio and accumulative distribution are presented to analyze the seismic signal.Finally,singular values obtained by singular value decomposition of the time frequency representation of a seismic signal are used as orthogonal transform features in order to directly decrease the dimensionality of feature space.