Research On Weak Seismic Signal Denoising Based On RCSST And CNN

Seismic exploration is an effective method to solve the problem of oil and gas exploration.It helps people find oil and gas resources underground by analyzing and observing the propagation law of artificially excited seismic waves in the ground.However,the seismic wave received by the detector often contains a large amount of interference noise,which seriously affects the identification of effective signals.At the same time,as conventional oil and gas fields have reached the end of exploitation,people are gradually turning their attention to unconventional oil and gas fields.Among the exploration signals obtained,the effective signal is weaker,the interference noise is stronger.And noise suppression of seismic signals have become more difficult,which brings great difficulties to the interpretation of seismic data.Therefore,suppressing the interference noise in seismic signals and improving the signal-to-noise ratio(SNR)of seismic signals are important links in seismic exploration signal processing.This paper focuses on the noise suppression of mountain seismic records and desert seismic records.According to the different characteristics of two kinds of seismic records,we adopt two kinds of denoising algorithms based on Shearlet transform.The effectiveness of the improved methods is verified by simulation and actual record processing results.The Hard Threshold Shearlet Transform(TST)takes advantage of hard-threshold and the features of Shearlet transform,which can effectively suppress random noise.However,it is easy to excessive restrain the coefficients of effective signals when using a fixed threshold to shrink the noisy signals.In addition,seismic signal after denoising will appear false cophase axis,which will affect the denoising effect.A recursive cycle spinning Shearlet tranform(RCSST)based on adaptive threshold algorithm is proposed in this paper.the noisy signals is decomposed into different subbands of frequency and orientation responses using recursive cyclic translation and shearet transform,which can effectively suppress random noise and enhance the continuity of valid signals.Then,we propose a brand new adaptive threshold to prevent the coefficients being killed excessively and protect the amplitude of the effective signals.Experimental results show that the new method exhibits betterperformance in random noise suppression and valid signal preservation than the conventional methods,Especially,the proposed method can handle seismic data with very low signal-to-noise ratio.In recent years,with the development of computer technology,convolutional neural networks(CNN)have been developed rapidly.As an advanced depth learning algorithm,CNN has made a breakthrough in image processing,speech recognition,and other fields.According to the characteristics of low SNR of desert seismic signals,a deep residual convolutional neural network for Shearlet transform(ST-CNN)model is proposed to suppress the interference noise in desert seismic signals in this paper.The model is divided into two phases,training phase and test phase.In test phase,The Shearlet coefficients of desert seismic signals are taken as the inputs of the model and the Shearlet coefficients of interference noise are used as the tags.The mapping relationship between inputs and tags are obtained by using CNN.In the later testing phase,we can use this mapping relation to predict the Shearlet coefficients of the disturbance noise from the desert seismic signals,then obtain the Shearlet coefficients of the effective signals.Finally,we can obtain the denoised desert seismic signals by Shearlet inverse transformation.In order to verify the effectiveness of this algorithm,we apply this algorithm to simulated and actual desert seismic records.By comparing the experimental results,it is found that ST-CNN is superior to the traditional denoising algorithm in the recovery of effective signals and the suppression of noise.