Seismic random noise suppression based on SP-DnCNN neural network
Seismic random noise suppression has been a persistent challenge in seismic data processing.It can directly determine the quality of successive seismic data processing and interpretation.Unlike coherent noise,random noise carries the characteristic of wide frequency band,which makes it hard to be separated from effective signals in the seismic data.To solve the above-mentioned problems,we propose to perform random noise suppression with the help of deep convolutional neural networks.Compared with conventional random noise suppression methods,methods based on neural network are more efficient and automatic.However,random noise suppression with existing deep learning algorithms ignores the existence of complex structure,resulting in energy loss of effective signals.In this manuscript,we propose a novel random noise suppression network,which is referred to as structure-preserving deep denoising convolutional neural network.The training labels in the manuscript are denoised field and synthetic seismic data,the features of which are used to train the neural networks.And considering the drawback of conventional neural networks in preserving the structure of seismic signals,we introduce local seismic dip information into the network and incorporate local seismic dip information into the objective function.Numerical and field seismic data applications indicate that the neural network can effectively suppress random noise in seismic data.And structure of seismic data is preserved with the introduction of local dipping information.
万方数据
维普
