Research On Seismic Signal Denoising And Reconstruction Based On Unet

In practical exploration,due to the influence of environmental or instrument factors,noise disturbance and seismic track loss often occur in seismic signal acquisition,which interferes with subsequent seismic data processing.Although the traditional method can suppress the noise and reconstruct the missing track to a certain extent,the processing result has some disadvantages such as low signal-to-noise ratio,damage effective signal and inaccurate detail processing.Therefore,it is necessary to come up with a more efficient and intelligent method.In recent years,algorithms represented by deep learning have been involved in many fields and achieved remarkable results.In particular,the excellent performance of convolutional neural network in image data processing has expanded ideas for exploring new methods of seismic signal processing,and the application of deep learning technology in this field will gradually become the mainstream.In this paper,based on deep learning theory,two algorithms are improved for seismic signal denoising and reconstruction.In order to solve the problems such as low signal-to-noise ratio and weak generalization ability of random noise removal in seismic signals,a convolutional neural network based on fusion residual attention mechanism is proposed in this paper.In this method,a large number of clean signals and noisy signal sample data sets are obtained by sliding data blocks and using data enhancement strategies.After three parts of network training including encoding,decoding and jump connection,most of the noise can be filtered out and effective signals can be protected as far as possible.The experimental simulation results show that,for synthetic and actual seismic signals,the noise reduction performance of the proposed method is significantly improved compared with the traditional method and Unet method,which indicates that the proposed method can improve the noise reduction effect and achieve the purpose of effectively suppressing random noise.In order to solve the problems such as low degree of texture detail restoration and poor robustness in the reconstruction of random missing seismic tracks in seismic signals,an improved U-shaped convolutional neural network based on random sampling is proposed in this paper.Similarly,this method generates a large number of missing signals and complete signal sample data sets through data block sliding and data enhancement strategies.After network training,the network learns and detects the location of missing tracks,and finally restores the gaps in seismic signal records as much as possible.Through experimental simulation and comparison,the proposed algorithm has excellent reconstruction performance in the synthetic seismic signal,and the profile after processing is closer to the complete data.In practical seismic signal applications,the processing results of this algorithm also show that it has a high degree of detail restoration and maintains good robustness.