摘要
针对采用全卷积神经网络去除地震数据随机噪声方法中遇到的计算量大、容易出现过拟合等问题,提出了一种基于LeNet-5改进的卷积神经网络对地震数据进行去噪的方法。除去输入层,该方法包含2个卷积层、2个池化层和1个全输出层。采用误差最小的实验试选法,首先在单层卷积网络中确定第1个卷积层和池化层的参数,基于第1层参数确定第2个卷积层和池化层的参数,最后采用12000个大小为32×32的地震数据训练LeNet-5,采用1000个相同大小、相同信噪比的地震数据测试系统。Marousi2叠前和叠后地震数据去噪实验均表明,本文方法对水平和倾斜同相轴地震数据的去噪效果较好。与奇异值分解算法、BP(Back Propagation)算法以及文献[9]中算法相比,本文方法能更好地去除噪声。
We propose an improved convolution neural network based on LeNet-5 to address the problems of large computation and over-fitting in the full convolution neural network based method for eliminating noise of seismic data.The network of the proposed method consists of two convolution layers,two pooling layers,and one full output layer,in addition to the input layer.By using the experimental selection method of minimum error,the parameters of the first convolution layer and the pooling layer in the single-layer convolution network are determined.Then the parameters of the second convolution layer and the pooling layer are determined based on the parameters of the first layer.Finally,12000 seismic data with size of 32×32 are used as inputs to train LeNet-5,and 1000 seismic data with the same size and signal-to-noise ratio are used for testing the system.Experiments on pre-stack and post-stack seismic data from Marousi 2 model demonstrate that the proposed method has good denoising effect for horizontal and inclined in-phase axis seismic data.Compared with the singular value decomposition algorithm,BP(back propagation)algorithm,and algorithm in Ref.[9],the proposed method has better denoising effect.
作者
崔少华
李素文
汪徐德
Cui Shaohua;Li Suwen;Wang Xude(College of Physics and Electronic Information,Huaibei Normal University,Huaibei,Anhui 235000,China)
出处
《激光与光电子学进展》
CSCD
北大核心
2020年第6期263-270,共8页
Laser & Optoelectronics Progress
基金
国家自然科学基金面上项目(41875040)
国家自然科学青年基金(11504121)
安徽省教育厅项目(2018jyxm0530,2017kfk044,KJ2017B008,201910373104)。
