Seismic Data Reconstruction Based On Compressed Sensing And Deep Learning

With the development of oil and gas exploration,the increasingly complex exploration environment usually leads to incomplete and irregular seismic data,which interferes with subsequent seismic data processing and ultimately affects the reliability of the processing results.Therefore,it is often necessary to reconstruct the data in the early stage of seismic data processing to restore the incomplete and irregular data acquired in the field into complete and regular data.This thesis firstly starts from the traditional model-driven seismic data reconstruction method,and studies the influence of different sparse bases on the reconstruction effect under the compressed sensing framework.It is found that the reconstruction effect is poor at low sampling rate and requires higher prior knowledge.In order to further improve the effect of seismic data reconstruction,this thesis combines the compressed sensing theory with the deep learning method,proposing a seismic data reconstruction method that combines model-driven and data-driven.The method firstly reconstructs the data through the compressed sensing method,then takes the initial reconstruction result as the input data,and the corresponding complete data as the label data.The convolutional neural network is used to realize further reconstruction,and finally obtains the seismic data reconstruction result with high precision and high signal-to-noise ratio.The effectiveness of the method in this thesis is verified by synthetic seismic data and actual seismic data.With the same random sampling rate,the reconstruction results of the combined compressed sensing and deep learning method can recover the missing data more effectively than those of the single compressed sensing or the deep learning method.The combined compressed sensing and deep learning method can show better reconstruction results even at lower sampling rates.At the same time,the combination of model-driven and data-driven processing in this thesis also provides a new idea for the high-precision reconstruction of seismic data.