Noise Attenuation Methods Of Seismic Data Based On Deep Learning

In the field of seismic data processing,suppressing noise and improving signalto-noise ratio of data(SNR)have always been the crucial parts.As we all know,there are many different types of noise in the raw seismic data,such as ground roll,random noise and multiple refraction wave.These noises often contaminate seismic signals and severely reduce the SNR of seismic data.According to the characteristics of noise in seismic data,the noise can be divided into two types: coherent noise and random noise.Among them,ground roll is a common type of coherent noise with high amplitude,strong spatial correlation,and low frequency.According to ground roll characteristic,a variety of methods for addressing ground roll attenuation have been developed.However,existing methods are limited,especially when using real land seismic data.For example,when ground roll and reflections overlap in the time or frequency domains,traditional methods cannot separate them,and often distort the signals during the suppression process.Compared with coherent noise,random noise has no fixed frequency and is often distributed in the whole frequency band,which is more difficult to denoise.Although there are many methods of random noise attenuation,when the subsurface structure becomes complex,traditional methods also cannot separate the random noise and reflections.Recently,with the development of deep learning,the deep neural network has also been used in the field of seismic exploration.Some progress has been made in fault detection,seismic facies classification,and reservoir prediction.Here,we use the denoising convolutional neural network and generative adversarial network algorithm to attenuate random noise and ground roll in seismic data,respectively.Unlike traditional methods for noise attenuation dependent on various filters,the neural network is based on a large training dataset that includes pairs of data with and without the ground roll.After training the neural network with the training data,the network can identify and filter out any noise in the data.This paper first analyzes the traditional noise suppression method to understand the shortcomings of traditional methods.Then,we describe the basic principle of neural networks and propose the denoising convolutional neural network and the generative adversarial network.Secondly,the principle of the two kinds of networks and how to apply them to the noise attenuation are explained in detail.Finally,tests on synthetic and real land seismic data show that the neural network algorithm is feasible for noise attenuation in seismic data.Compared with the traditional methods,the neural network algorithm has a significant improvement in noise suppression ability and protection signal.