Research On Noise Reduction Method Of Seismic Co-band Noise Based On Deep Learning

During the acquisition and propagation process of seismic signals,they are easily polluted by various random noises from geological conditions,surrounding environment and acquisition instruments.Among them,the same frequency band noise brings challenges to seismic signal noise reduction because of its non-stationary,high energy and the same frequency band as the seismic signal.The traditional seismic cofrequency noise reduction method has problems such as poor adaptive ability,manual selection of parameters and limited noise reduction ability,which makes the signal-tonoise ratio of noise reduction low.In recent years,with the continuous development of deep learning,its powerful feature learning ability,high efficiency,accuracy and high adaptability are beneficial to deal with the same frequency band noise in seismic signals and improve the signal-to-noise ratio of seismic signals.Therefore,this paper uses the deep learning method to study how to remove the same frequency band noise in the seismic signal and improve the Signal-to-Noise Ratio of the seismic signal.The main research contents are as follows:(1)In view of the problems that the traditional seismic signal noise reduction method has low noise reduction ability and low Signal-to-Noise Ratio in the same frequency band,taking advantage of the multi-scale feature extraction of U-shaped convolutional neural network(U-Net),combined with Residual Dense Blocks,a noise reduction model Residual Dense Block U-shaped Network(RDBU)is proposed in the same frequency band of earthquakes.The model uses Residual Dense Blocks to replace the ordinary convolutional layers in the U-Net,thereby enhancing the feature learning ability of the network and improving the noise reduction performance of the model.In order to test the noise reduction effect of the RDBU model,the RDBU model uses the Stanford Global Earthquake dataset to construct training and test sets,and conduct training and testing.The test results are compared with the label signal,and the Signal-to-Noise Ratio(SNR),Correlation Coefficient(r)and Root Mean Square Error(RMSE)are calculated to evaluate the noise reduction effect of the model.The experimental results show that compared with the traditional method,the Signal-to-Noise Ratio and Correlation Coefficient of the RDBU model are increased by6.52 d B and 26.43%,respectively,and the Root Mean Square Error is reduced by52.41%.It shows that the RDBU model has strong noise reduction ability and can significantly improve the Signal-to-Noise Ratio of seismic signals.(2)Although the RDBU model has a strong ability to reduce the noise in the same frequency band of the earthquake,the seismic signal after noise reduction still has the problem of waveform distortion,and the noise reduction Signal-to-Noise Ratio is still not ideal.Aiming at this problem,based on the improvement of the RDBU model,combined with the Atrous Convolution,a seismic co-band noise reduction model ARDU(Atrous Residual Dense Block U-shaped Network)is proposed.The Atrous Convolution in the model can expand the receptive field,extract more effective signal features,reduce waveform distortion,and improve the Signal-to-Noise Ratio of seismic signals without increasing network parameters.To demonstrate the effectiveness and feasibility of the ARDU denoising model,the ARDU model is trained and tested using the same dataset as in(1),and its test results are compared with the label signal.The experimental results show that compared with the RDBU model,the Signal-to-Noise Ratio and Correlation Coefficient of the ARDU model are improved by 0.8d B and1.67%,respectively,and the Root Mean Square Error is reduced by 6.37%.This shows that the designed ARDU model can alleviate the waveform distortion of seismic signals to a certain extent,further improve the Signal-to-Noise Ratio of seismic signals,and improve the quality of seismic signals.(3)In order to further test the noise reduction performance of ARDU model on actual seismic signals,the noise reduction test experiments were carried out using the actual seismic signals of Wenchuan and Jiuzhaigou,and the Signal-to-Noise Ratio estimation algorithm was used to calculate the Signal-to-Noise Ratio of the actual seismic signals before and after noise reduction.The experimental results show that because the noise of the actual seismic signal is more complicated than that of the simulated seismic signal,the noise reduction Signal-to-Noise Ratio of the model decreases slightly.Among them,the improvement of the Signal-to-Noise Ratio of the ARDU model to the actual seismic signal is reduced by 17.01% compared with the simulated signal.However,compared with other noise reduction methods,the improvement of the Signal-to-Noise Ratio of the ARDU model is still optimal,and the noise reduction performance can meet the needs of practical applications.