To reduce the cost of seismic acquisition,improve the acquisition efficiency and maintain the regularity and completeness of seismic data,the regular projection of geometry is proposed to fill missing shots and receivers for random seismic data acquisi-tion.Under the framework of compressive sensing,dictionary learning and sparse representation are used to reconstruct seismic da-ta.Compared with traditional dictionary learning method,the proposed method replaces orthogonal matching pursuit(OMP)with batch-OMP to avoid heavy computation in direct inversion of matrix,and also uses alternating least squares(ALS)to take place of singular value decomposition(SVD)to improve computation efficiency.Moreover,to avoid fitting noise,a damped constrain is ap-plied to sparse coefficients for obtaining better dictionary atoms.Frequency domain dictionary and seismic data reconstruction are proposed to tackle the issues of low computational efficiency and poor capability of protecting weak signal in the time domain using traditional dictionary learning methods.Only seismic data in principal frequency band is used to reconstruct seismic data,which can effectively reduce computation workload,suppress noise and improve signal-to-noise ratio.Thus,the technical process of regular projection of geometry and seismic data reconstruction for random acquisition of seismic data is formed.The proposed method is applied to field data,demonstrating that the quality of prestack seismic data is effectively improved through regular projection of geometry and seismic data reconstruction,contributing to better imaging results.
dictionary learningsparse representationdamped constrainregular projection of geometryseismic data reconstruction
维普
万方数据
