Research On Seismic Signal Denoising Based On Patch Structural Similarity

Seismic signal denoising is particularly critical to subsequent seismic signal processing.The use of patch structure similarity of seismic signals helps denoising algorithm to identify effective signals and noise more accurately.By dividing the seismic signal into patchs,using the similarity and redundant information between the signal patchs,the correlation of the effective signal is enhanced,which lays the foundation for the seismic signal denoising algorithms.In this thesis,the research on seismic signal denoising based on patch structure similarity is proposed.The main contents are as follows:(1)Convolutional neural networks(CNN)for seismic signal random noise attenuationA denoising algorithm based on CNN for seismic signal random noise attenuation is proposed.The key of the algorithm is to build a CNN for seismic signal denoising,which mainly includes the input layer,convolution layers,activation layers,normalization layers,and the output layer.In the process of training,CNN reads the signal matrix composed of seismic signal patchs without the noise in the input layer,extracts and processes the seismic signal via the multiple convolution layers,extracts the fluctuation characteristics of the signal using the Rectified Linear Units in the activation layer,accelerates the training convergence using the normalization layer,and finally obtains the random noise as the output in the network output layer by residual learning.While,in the process of CNN denoising,the complete noisy seismic signal is as the input layer.The experiments on the real marine pre-stack seismic signal,the post-stack land seismic signal,and the complex land post-stack seismic signal illustrate that the CNN is feasible in denoising.In addition,by comparing with some common denoising algorithms,such as wavelet transform,dual-tree complex wavelet transform,and curvelet transform,the proposed algorithm demonstrate stronger denoising capability,and has certain reference value for exploring new methods of seismic signal denoising.(2)Seismic signal blind denoising based on W weighted kernel norm minimization(W-WNNM)A denoising algorithm for blind seismic signal based on W-WNNM is proposed.First,use noise level estimation algorithm based on the weak texture patch selection(WPCA)to estimate the noise level of noisy seismic signal,that is,divide the noise seismic signal into patchs,in the iterative process,filter the weak texture signal patchs with the updating threshold value,and then estimate the noise level through principal component analysis to get the final estimation;Then take the estimated value as the input parameter of the weighted kernel norm minimization(WNNM)to remove random noise,which controls the contraction degree of matrix singular value with the weight distribution in the iterative process to approach original seismic signal.The experiments on the real marine pre-stackseismic signal,the landshot seismic signal and the post-stack land signal demonstrate that the W-WNNM without known noise level can effectively remove the noise of seismic signal.In addition,by comparing with some common denoising algorithms,such as,dual tree complex wavelet transform,curvelet transform,WNNM.The proposed algorithm illustrates stronger denoising capability and higher denoising efficiency,and has certain reference value for exploring the seismic signal blind denoising.(3)Seismic signal denoising based on weighted Schatten p-norm minimization(WSNM)WSNM is used in seismic signal denoising.Compared with the traditional trace norm,WSNM can guarantee more accurate signal repair.It is inspired by Schatten p-norm and WNNM that divide the seismic signal into patchs to construct the low rank matrix combining with the patch structure similarity of seismic signal to finish the denoising with WSNM.WSNM and WNNM are used to denoise the noise of the real marine pre-stack seismic signal,the landshot seismic signal and the post-stack land signal respectively.In the experiments,the optimum value of parameterpof WSNM is selected to compare the denoising performance with WNNM.The experimental results show that the proposed algorithm is more effective than WNNM in seismic signal denoising,which verifies the feasibility and effectiveness of the algorithm,and has certain reference value to the research on seismic signal denoising at high noise level.