(吉林大学 仪器科学与电气工程学院，长春 130026)
摘 要: 时间域航空电磁数据经预处理后，存在残余噪声，严重影响晚期道数据质量。提出一种基于主成分分析的去噪方法，通过对航空电磁剖面数据的自相关矩阵进行奇异值分解，得到特征值及对应的特征向量，将电磁数据通过旋转矩阵（特征向量矩阵的转置）进行线性映射，得到其主成分。大特征值对应的低阶主成分反映相关性较强的电磁信号，而小特征值对应的高阶主成分反映不相关噪声。采用低阶主成分与特征向量重构电磁数据能够去除不相关噪声。采用主成分分析法分别进行了仿真数据与实测数据的主成分特征分析与噪声去除。仿真数据去噪后，信噪比提高了13 dB；而野外飞行实测数据的噪声水平也由±25 nT/s 降低到±8 nT/s。
关键字: 时间域航空电磁；主成分分析；特征值；特征向量；去噪
（School of Instrumentation and Electrical Engineering, Jilin University, Changchun 130026, China）
Abstract:There is still residual noise which affects the quality of later channel data after preprocessing for time domain airborne electromagnetic data. An approach was proposed to remove the residual noise based on principal component analysis. The principal components were computed through the rotation matrix which is the transpose of eigenvectors matrix. The low-order principle components associated with the big eigenvalues reflect the correlated electromagnetic signals, while high-order principle components associated with the small eigenvalues are corresponding to the uncorrelated the noise. Therefore, the electromagnetic data are reconstructed by suitable number of the low-order components to remove uncorrelated noise. The experimental results of the simulation data show that the SNR is improved of 13 dB. The peak to peak value of the latest two channels for the field survey profile data is reduced from ±25 nT/s to ±8 nT/s after noise removal.
Key words: time domain airborne electromagnetic method; principal component analysis; eigenvalue; eigenvector; noise removal
