Denoising of Seismic Data Based on DnCNN and Multi-Scale Feature Extraction
Seismic data are often mixed with various noises,which seriously affect the interpretation and application of the data.Tradi-tional dienoising methods often have many limitations in processing seismic data and cannot meet the requirements of accurate denois-ing.A seismic data denoising method based on Denoising-CNN(DnCNN)and multi-scale feature extraction is proposed to overcome these limitations.Channel attention mechanism,spatial attention mechanism and depth separable convolution are added into the multi-scale feature extraction module,and by using the residual network structure for reference,the network can not only learn features of dif-ferent scales,but also reasonably allocate weights of different channels and spaces,and fully utilize the correlation between data.This method not only significantly improves the effectiveness of network training,but also preserves the effective signals and local details of the original seismic data to the greatest extent possible while denoising.
seismic data denoisingdeep learningmulti-scale featureDnCNN
万方数据
维普
