基于特征提取与聚类识别的人工源电磁伪随机信号处理方法

人工源电磁数据易受噪声干扰，影响勘探效果。传统人工源电磁数据处理通常采用频点筛选、异常剔除等方法，人为因素影响太大，且滤波方法无法保留伪随机有效信号。为解决这些问题，针对人工源电磁伪随机数据，通过剖析有用信号与噪声的时域特征，定量辨识并定性分析人工源电磁伪随机有用信号，提出了基于特征提取和聚类识别的人工源电磁伪随机信号处理方法。首先，建立两类典型噪声和伪随机信号的样本库，分析样本库信号的时、频域特征；然后，提取时域统计学特征，并结合模糊C均值聚类算法识别并去除噪声，保留有用信号并重构人工源电磁原始数据；最后，利用数字相干技术提取有效频点的频谱。对模拟数据与实测数据进行处理分析，结果表明：本方法能准确、有效地识别并剔除典型噪声，显著提高人工源电磁伪随机数据的质量，经本文方法处理后的电场分量E_x归一化电场曲线和广域视电阻率曲线更平稳、连续，可有效提高人工源电磁信号的信噪比。

CSEM pseudo-random signal processing method based on feature extraction and clustering identification

The controlled-source electromagnetic (CSEM) data is susceptible to noise,which leads to unsatisfactory exploration effects. Human factors will exert a huge influence on traditional CSEM data processing which usually employs frequency point screening,abnormal elimination and other methods and the filtering method cannot retain pseudo-random effective signals. According to the recorded CSEM data in the time domain,we analyze the time-domain statistical characteristics of useful signals and noises in the CSEM data,and quantitatively identify and qualitatively analyze useful CSEM signals to address the above problems. As a resulta CSEM pseudo-random signal processing method based on feature extraction and clustering identification is proposed in this paper. Firstly,the sample library including two kinds of typical noises and pseudo-random signals is established,and features of the time and frequency domains of the sample library signals are analyzed. Then the time-domain statistical features are extracted,and the fuzzy C-means clustering algorithm is adopted to identify and eliminate the noise for retaining useful signals and reconstructing original CSEM data. Finally,the frequency spectrum of effective frequency points is extracted by digital coherence technology. Through the processing of simulated data and measured data,results show that the proposed method can identify and eliminate typical noises accurately and effectively,thereby significantly improving the quality of CSEM data. After being processed by the proposed method,the component E_x normalization electric field curve and wide field electromagnetic (WFEM) resistivity curve are smoother and more continuous, thus effectively increasing the signal-to-noise ratio of CSEM signals.