Research On Seismic Signal Denoising Method Based On Sparse Coding And Dictionary Reconstruction

As oil and gas exploration deepens into complex oil and gas reservoirs,seismic data collected in the field affected by geological conditions often suffers from severe noise interference,resulting in low signal-to-noise ratio and resulting attenuation of effective signal energy.These factors are not conducive to field data for geological interpretation.However,the digital processing of seismic data requires seismic data with high signal-tonoise ratio,high resolution,and high fidelity.At the same time,improving the data signalto-noise ratio is also a top priority for seismic data processing.On the other hand,the continuous expansion of the seismic exploration area,the seismic data collected in the field has a large amount of data and rich information,and the sparse representation theory based on compressed sensing has the ability to automatically extract the main features of the data and express the data features well.It can effectively reduce the dimension of seismic data,and use effective low-dimensional features to represent massive highdimensional seismic data.In the sparse representation principle part,this article mainly introduces the sparse solving algorithm and dictionary learning algorithm.Among the sparse decomposition algorithms,we mainly study the greedy algorithm for solving 0-norm and the convex optimization algorithm for solving 1-norm.Dictionary learning algorithm mainly introduces DCT dictionary,MOD dictionary and KSVD dictionary learning algorithm.The method of super-complete dictionary learning based on sparse representation theory to remove random noise from seismic data is studied,and the above three methods are tested on the synthesized seismic data.The learning dictionary is better than the fixed DCT dictionary,because the learning KSVD dictionary uses a singular value decomposition algorithm to update the dictionary atoms column by column,it has a better denoising effect than the MOD dictionary method.Since the KSVD dictionary is not suitable for processing large-scale seismic data,this article introduces an online dictionary learning method suitable for large-scale processing of pictures.The algorithm is applied to seismic data denoising and the ADMM algorithm is used to replace the LARS algorithm in the online dictionary learning algorithm to solve sparse coding,and the synthetic and actual seismic data are tested to effectively remove random noise and retain effective signals.