中文摘要：

为了剖析大地电磁信号和强干扰的本质特征，进一步精细分离出微弱的大地电磁有用信号，提出基于递归分析和聚类的大地电磁信噪辨识及分离方法.首先，运用递归分析法扩展大地电磁一维时间序列的维数，分析了嵌入维数、延迟时间和判别阈值对递归图的性能，并研究了不同长度的序列对递归定量分析参数的影响情况.然后，构建典型的大地电磁强干扰类型和微弱的大地电磁有用信号样本库，针对样本库讨论了强干扰和微弱大地电磁信号之间的递归定量分析参数，分析了K均值聚类和模糊C均值聚类的信噪辨识效果.最后，对实测大地电磁数据进行信噪辨识处理，并仅对辨识为强干扰的时间段采用数学形态滤波进行噪声压制.实验结果表明，递归分析能定性及定量地描述大地电磁信号时间序列的非线性特征和原动力系统的本质规律，与聚类算法相结合能对矿集区实测大地电磁信号进行信噪辨识；处理后的卡尼亚电阻率-相位曲线更为光滑、连续，其结果更为精细地保留了大地电磁信号低频段的缓变化信息，整个低频段的大地电磁数据质量得到了明显改善.

英文摘要：

In order to analyze the essential characteristics of magnetotelluric （MT） sounding data and strong interferences, and further to separate weak useful MT signal, we propose a new method for identification and separation of MT signal and noise based on recurrence analysis and clustering. First, we use the recurrence analysis method to extend the dimension of MT time series and analyze the embedding dimension, delay time and determination threshold, and study the Recurrence Quantification Analysis （RQA） parameters for the sequence of different lengths. Then, we build a sample dataset for typical strong interferences and weak useful MT signal. According to the sample dataset, we discuss the RQA parameters between strong interferences and weak MT signal, and analyze the effects of signal and noise identification for the K-means clustering and fuzzy C-means clustering. Finally, the real MT data is processed through signal and noise identification, and only the time series containing strong interferences are suppressed by mathematical morphology filtering. Experimental results show that the recurrence analysis can qualitatively and quantitatively describe the nonlinear characteristics of the time series and the essential rule of the prime power system for MT. Combined with clustering algorithm permits to identify the measured MT data in ore concentration area. The Cagniard resistivity-phase curve is more smooth and continuous after using the proposed method. Moreover, the slow change information of low frequency for MT is more retained finely, and the quality of MT data for the overall low frequency band is improved significantly.