Adaptive De-noising Of Magnetotelluric Data Based On SVM And MMF-KSVD Dictionary Learning

MT method is widely used in the fields of deep earth structure exploration,mineral resources exploration,geothermal exploration.Because of the natural electromagnetic source,the MT signal is susceptible to cultural noise interference.The conventional MT time domain denoising method tends to lose the effective low-frequency signal and part of the high-quality signal when there is persistent human noise interference.To solve the problem,we use SVM and dictionary learning in machine learning for MT strong noise interference suppression,combined with MMF to extract low-frequency signals that can prevent the loss of low-frequency signals.On the basis of SVM pattern recognition,identify high quality signal fragments and noisy signal fragments,and use K-SVD dictionary learning to denoise the noisy signal fragments,and extract the noise contours by obtaining the feature structure of the noise through autonomous learning from the observed data,so as to achieve signal-noise separation.The main results achieved in the thesis are:(1)We use statistical parameters such as Ap En,Samp En,Fuzzy En and Box D to evaluate the quality of MT data,as there are large differences in the curve patterns of highquality signals and noisy signals in the observed data,based on the signal sample library,constructing separation hyperplane of SVM and identify signal classes autonomously.SVM has good recognition effect of typical noise interference.(2)The organic combination of MMF and K-SVD dictionary learning has a certain suppression effect on the noise in MT,and the denoising effect is better than Wavelet,and the apparent resistivity-phase curve of the data is greatly improved.(3)Combining MMF,SVM and K-SVD dictionary learning to process the observation data,the typical noise suppression effect is obvious,and the signal-noise ratio is improved substantially.The observation data of different work areas,after processing,the data quality is greatly improved,and the method is better than Robust,which can present more fine structure information and improve the deep resolution.The MT adaptive denoising method based on SVM and MMF-KSVD dictionary learning effectively eliminates various strong interference,and improves the quality of MT data,better restores the original characteristics of MT data.