基于卷积神经网络的海洋地震勘探拖缆混采数据分离

Deblending of the Simultaneous Sources Acquisition Streamer Data Based on CNN

近年来,混合震源采集技术在海洋勘探中发展迅速,由于其高效的采集方式,以及能够获取高质量的地震数据而受到青睐。对混合震源数据进行有效的混采分离是混合震源采集技术成败的关键,本文结合近年来发展火热的卷积神经网络(CNN)方法并根据地震资料的特点,提出了一种基于神经网络模型的数据驱动混采分离方法。通过制作的两万余个样本的混采数据集进行训练,获得混采分离模型,通过测试集的试验以及与常规混采分离方法进行对比,结果表明本文采用的CNN混采分离方法具有较高的分离精度,且效率很高。

In the recent years,the simultaneous sources acquisition technology has been developing rapidly in Marine exploration,which owns efficient collection and could obtain high quality seismic data.The key of the simultaneous sources acquisition technology is deblending the simultaneous sources data effectively.This paper intergrated the convolutional neural network method developed in recent years to propose a data-driven deblending method according to the neural network model,which is constructed according to the characteristics of seismic data.And a simultaneous sources acquisition data set containing more than 20000 samples is made for training the deblending model.The result of CNN method has excellent separation accuracy and high efficiency according to the test set and the comparison with the conventional deblending method.