Strong Noise Suppression For Magnetotelluric Data With Adaptive Sparse Representation And NPSO-OMP

The magnetotelluric(MT)method employs natural electromagnetic fields as sources,which is a popular and indispensable method in the field of deep earth exploration.However,natural MT signals are weak,non-stationary and random,which are therefore susceptible to human noise.In this paper,the self-organized competitive neural network(SCNN)combined with the niche particle swarm optimization(NPSO)-orthogonal matching pursuit(OMP)sparse decomposition method are used to suppress the strong electromagnetic interference in the ground.The main research work are as follows:(1)The time domain characteristics of high-quality and noisy magnetotelluric data are systematically analyzed.Magnetotelluric time series are cut into segments with appropriate length and marked as high-quality or noisy samples(each segment is a sample).Then calculate the peak,root mean square,variance,standard deviation,sample entropy,fuzzy entropy and other characteristic parameters of each sample.It shows that high-quality samples have significantly different eigenvalues than samples with low signal-to-noise ratio.(2)The self-organizing competition network algorithm is used instead of manual discrimination to realize automatic signal-to-noise identification.Manual discrimination requires operator with rich experience,which not only results in a large workload,but also has different standards for different operator.To this end,an automatic discrimination method based on self-organized competitive network neural is developed.The validity of the method is verified by case study of simulated and measured data.(3)Aiming at the impulsive-like noise often appearing in magnetotelluric data,a redundant dictionary with good adaptability was constructed,and the strong noise extraction was combined with the orthogonal matching pursuit algorithm.Aiming at the problems of low efficiency and time consuming of OMP algorithm,NPSO algorithm was used to optimize the OMP decomposition process.Simulation experiments and measured data processing results show that compared with conventional methods such as wavelet and EMD,the NPSO-OMP method can more accurately separate large-scale interference and better retain effective signals.(4)Combining signal-noise identification of self-organizing competitive network neural with NPSO-OMP sparse decomposition,an adaptive denoising method based on sparsity is proposed.The maximum variance of the high-quality signal identified by the self-organizing competitive network is used as the stopping condition for the sparse decomposition,instead of manually setting the stopping condition,the automatic stopping of the NPSO-OMP sparse decomposition is realized.This paper proposes an automated noise attenuation method for magnetotelluric data using SCNN and NPSO-OMP.Simulated experiments and measured data processing show that the newly proposed method can accurately identify the noisy fragments,and effectively isolate the strong human noise among them,which significantly improving the quality of magnetotelluric data.