Seismic Signal Denoising Based On Sparse Transform And Statistical Nearest Neighbor

Due to factors such as the acquisition environment and the control of acquisition costs,seismic signals are inevitably doped with different levels of random noise during the acquisition process.Random noise can seriously affect the subsequent processing steps such as horizontal stack,inversion and migration.Therefore,seismic signal denoising has important research significance.Wavelet(WT)and Contourlet(CT)and other sparse transform methods have been widely used in the field of signal denoising.Non-Local Means(NLM)and Bilateral Filter(BF)are also classic denoising methods,and the Statistical Nearest Neighbors(SNN)neighbors selection strategy can improve the denoising performance of NLM and BF.In order to separate seismic signals and noise more effectively,this paper studied the denoising of seismic signals based on sparse transform and SNN.The main research contents are as follows:(1)Seismic signal denoising with Wavelet-Contourlet(WCT)based on cycle spinningIn order to carry out effective signal-to-noise separation algorithms,we combined the WT and CT,and eliminated the pseudo-Gibbs phenomenon by cycle spinning.In this paper,a new WCT algorithm with cycle spinning of seismic signal denoising(CS-WCT)was proposed.The algorithm first uses cycle spinning to shift the seismic signal,then replaces the Laplace pyramid(LP)transform in CT with WT.And then uses the threshold function to remove the noise in the signal.After inverse transformation and inverse cycle spinning,we obtain the denoised seismic signal.Denoising experiments with synthetic seismic signals,post-stack land signals and pre-stack marine seismic signals,and compared the results with WT,CT and WCT.The experiment chose Mean Square Error(MSE),Peak Signal-to-Noise Ratio(PSNR)and Structural Similarity(SSIM)as evaluation indexes.The results showed that the proposed algorithm has the best denoising effect.(2)Seismic signal denoising with NLM based on SNNIn order to carry out effective signal-to-noise separation algorithms,the Nearest Neighbors(NN)neighbor sets selection strategy can alleviate the computational burden of NLM.But at the same time,the prediction error will be introduced to affect the denoising performance.SNN neighbor sets selection strategy can reduce the above deviations and remove visual artifacts in the denoised signal.Therefore,we proposed a non-local means seismic signal denoising algorithm based on SNN.SNN collects neighbors with target block square distance close to their expected value as neighbor sets,and replaces NN method to determine the denoising neighbor sets with the closest distance to the target sub-block.Then the neighbor sets will be weighted and averaged to obtain the data points' value of the target sub-block after denoising.Repeat the same calculation for all the noisy target sub-blocks in the signal,and finally all the denoised target sub-blocks are aggregated into the denoised signal.Denoising experiments with synthetic seismic signals,post-stack seismic signals and pre-stack marine seismic signals,and compared the results with NLM and NN-NLM.The experiment chose MSE,PSNR and SSIM as evaluation index.The results showed that the proposed algorithm has the best denoising effect and can eliminate artificial artifacts.(3)Seismic signal denoising with BF based on SNNBased on SNN,this paper proposed a SNN based BF denoising algorithm(SNN-BF),which combined the SNN neighbors selection strategy with BF to improve the denoising performance.Similar to SNN-NLM,the algorithm determines neighbor sets by SNN method,then weights the neighbor sets to obtain average value and denoised estimate data points' value.Then the same calculation is performed for all data points of the noisy signal.Finally,the denoised signal is synthesized by all estimate data points.Denoising experiments with synthetic seismic signals,post-stack land seismic signals and pre-stack marine seismic signals,and compared the results with NN-BF.MSE,PSNR and SSIM were used to evaluate the indicators.The results showed the proposed algorithm denoising effect.