Analysis of the effect of using batch normalization layers in convolutional neural networks on seismic data denoising
Deep learning algorithms have been widely applied in seismic data processing,and have achieved many good applications in seismic data denoising and other related domains.Current research primarily focuses on selecting and applying different deep learning algorithms,network structures,and labeling methods.However,less attention is paid to the impact of inherent dataset variations on the application of deep learning algorithms.This paper analyzes the impact of Batch Normalization in Convolutional Neural Networks(CNN)on seismic data denoising.By employing theoretical formulas and conducting comparative numerical calculations,this study proposes recommendations for utilizing batch normalization layers based on the analysis of seismic data features.The suitability of incorporating batch normalization layers relies on the statistical distribution characteristics of the dataset.Effective improvements in network performance can be achieved only when the normalized energy distribution of the training set is concentrated within a strong amplitude region.Nevertheless,in seismic data denoising,it is generally advised to refrain from using batch normalization layers.These findings offer valuable insights for the improved application of deep learning algorithms in seismic data denoising.
Deep learningBatch normalizationStructure optimizationDenoisingData characteristics analysis
万方数据
维普
