Research On The Algorithm Of Seismic Signal Noise Reduction Based On Deep Learning

Earthquake usually causes a lot of casualties and property losses.Earthquake warning is an effective means to reduce disasters.Accurate seismic data can help the earthquake early warning system to get accurate time.However,due to the increasingly complex monitoring environment of stations,the seismic data collected will be strongly disturbed by various kinds of noise.The presence of noise in seismic data will have a great impact on the detection and seismic phase acquisition of the earthquake time,which will affect the accuracy of earthquake warning.Therefore,it is necessary to reduce noise of the seismic data.However,the noise reduction results of traditional seismic signal denoising algorithm are greatly affected by the selected feature function or threshold parameters,which leads to low signal-to-noise ratio(SNR)and waveform distortion.In recent years,with the development of deep learning,many scholars have tried to apply the deep learning method to the noise reduction of seismic signals.In order to further improve the application of deep learning in the field of seismic signal denoising,this paper focuses on the problem of seismic signal denoising based on convolution neural network,which can improve the SNR of seismic signal and reduce waveform distortion.The main contents of this paper are as follows:(1)Aiming at the problems of low SNR of seismic signals and waveform distortion in current deep learning denoising methods,a seismic signal denoising model based on DnCNN network is proposed.The DnCNN model can use the residual learning method to reduce the noise of seismic signals with different SNR without manual intervention.In order to test the noise reduction effect of the DnCNN model,the DnCNN network was trained and tested using the Stanford University global seismic data set,and its noise reduction results were compared with traditional noise reduction methods and deep learning methods.The experimental results show that the DnCNN model is compared to Traditional noise reduction algorithms and deep learning methods can achieve better noise reduction effects and improve the SNR of seismic signals.(2)Although DnCNN model can improve the SNR of seismic signals to a certain extent,the utilization rate of feature maps is not high,and some details are easily lost,which leads to a certain degree of waveform distortion of the noise reduced signals.In order to solve this problem,a convolutional depth network noise reduction model DnRDB is proposed,which combines the residual density block(RDB).The RDB block in the model uses dense connection and feature fusion to increase the utilization of feature map,improve the SNR of seismic signal and reduce waveform distortion.In order to prove the noise reduction effect of DnRDB model,the same data set is used to train and test the DnRDB model.The experimental results show that DnRDB model can effectively remove all kinds of noise in the test set.Comparing DnRDB model with DnCNN model and other deep learning models,after denoising the same segment of noisy seismic signal,the average SNR is increased by about 0.7d B and 0.12 d B respectively,and the average correlation coefficient is increased by about 0.0573 and0.0093 respectively.The DnRDB model has the best denoising effect,which can effectively improve the SNR of seismic signals and the minimum distortion of waveform.(3)Noise in the seismic signal will affect the accuracy of seismic phase identification,and noise reduction of the seismic signal containing noise can improve the effect of seismic phase identification.In order to test the noise reduction effect of the DnRDB model in actual seismic signals,the DnRDB model was applied to the 2008 Wenchuan earthquake aftershock data set to reduce the noise of the seismic signals in the data set.Use the long and short window seismic phase recognition algorithm to identify the seismic signal before and after noise reduction,and use the correct rate and recall rate to evaluate the effect of seismic phase recognition.The experimental results show that the Pg and Sg after noise reduction the accuracy of recognition is increased by 18% and 9%,and the recall rate is increased by 32% and 29%,respectively.Moreover,the accuracy of seismic phase recognition after noise reduction by DnRDB model is the highest compared with other noise reduction algorithms.The DnRDB model can effectively improve the recognition accuracy of the seismic phase recognition algorithm through noise reduction.