Deep Learning-based Coherent Noise Attenuation Methods Of Marine Seismic Data

Sufficient energy reserve is an important prerequisite for the development of a country.The fact that fossil fuels dominate the international energy market has never been changed and seismic technology has always been the most important means to explore geological resources.China is in an important historical stage of removal with the demand for oil and gas resources is steadily increasing,leading to a gradual shift in emphasis from onshore to offshore seismic exploration.Given this situation,it is urgent to enlarge the scope of marine seismic exploration while developing more rapid and economical acquisition technology and data processing methods.According to the general classification principle,marine seismic noise can be divided into two categories: random noise and coherent noise.Nowadays the attenuation of random seismic noise is not a real problem for the industry.In contrast,coherent noise is the most harmful type of noise that can reduce the signal-to-noise ratio of marine seismic data and affect the quality of seismic imaging.Blending noise in blended data has the strongest coherence in all the marine coherent noise.Its kinematics and dynamics characteristics are almost consistent with the target signals.Another common and hard-to-attenuate type is seismic interference noise which occurs when energy from nearby marine seismic source vessels is recorded during a seismic survey.It kinematically presented as linear or with curvature.The frequency band range of seismic interference noise is equivalent to that of the source wavelet,and the amplitude level is within the same range as the useful signal,causing coherent artifacts in the recorded data.Marine coherent noise significantly affects the accuracy of seismic imaging and even misleads the conclusion of geological analysis.Over the years,the industry has developed various denoising techniques for seismic interference removal,but although well performing they are still time-consuming in use.Deep learning-based processing represents an alternative approach,which may significantly improve the computational efficiency.In case of conventional images,artificial neural networks are frequently employed for denoising purposes.However,due to the special characteristics of seismic data as well as the noise,they always fail in the case of marine coherent noise.We therefore propose two deep convolutional neural networks for the attenuation of two mentioned types of marine coherent noise.To secure a realistic study,only seismic field data were employed,both including more than 20000 training examples respectively.A deep convolutional neural network is firstly proposed to attenuate blending noise.The blended data are sorted from the common source to the common channel domain to transform the character of the blending noise from coherent events to incoherent contributions.After training and validating the network,seismic deblending can be performed in near real time.Experiments also show that the initial signal-to-noise ratio is the major factor controlling the quality of the final deblended result.The network is also demonstrated to be robust and adaptive by using the trained model to firstly deblend a new data set from a different geological area with a slightly different delay time setting,and secondly to deblend shots with blending noise in the top part of the record.Finally,a comparison shows it performs deblending with results comparable to those obtained with conventional industry deblending algorithms and much faster.A customized U-Net is proposed to attenuate seismic interference noise.Different from standard U-Net,the customized U-Net has fewer convolutional layers and adopts larger convolutional kernel in the first three convolution layers.It uses element-wise summation as part of the skip-connection blocks to ensure information fusion between high-and low-level features and to handle the vanishing gradient problem.Considering the uncontrollability of sources from other companies causing seismic interference noise,we train the customized U-Net in common shot domain with various noise from different directions and different distances.The customized U-Net was found to perform well except for the case when seismic interference noise comes from the side.We further demonstrate that such noise can be treated by slightly increasing the depth of our network.The attenuation results of data from another geological area without retraining demonstrate that the network is robustness.Although our customized U-Net does not outperform a standard commercial algorithm in quality,it can read and process one single shot gather in real time.This is significantly faster than any existing industry denoising algorithm,making this approach suitable for quality control onboard seismic vessels during acquisition.