物探化探计算技术2025，Vol.47Issue(3)：427-435,9.DOI:10.12474/wthtjs.20240923-0002
基于卷积自编码器的地震数据随机噪声压制
Random noise suppression of seismic data based on convolutional auto-encoder
摘要
Abstract
With the deepening of oil and gas exploration into complex geological environment,seismic exploration data are faced with more and morea complex geological environment,seismic exploration data are faced with increasingly serious severe noise interference,which poses a challenge to the subsequent data processing and interpretation.Because of the ReLU activation function used in conventional convolutional autoencoder denoising methods,there are problems of gradient disappearance and"dead ReLU"gradient disappearance and"dead ReLU"are problems.Therefore,in the process of deconnoisingvolution based on a convolutional autoencoder,ELU activation function is used to effectively avoid the problems of ReLU activation functionavoid the issues of the ReLU activation function,and the Adam optimizer is used to improve the stability and efficiency of the network.Firstly,the synthetic data is trained,and then the synthetic seismic data with noise and the actual seismic data are denoised.The results show that the autoencoder network constructed by ELU is superior to the original ReLU autoencoder in removing random noise,and can effectively separate effective signals from noise,significantly reduce the need for manual intervention,and realize intelligent noise reduction processing of seismic data,and improve the processing quality.关键词
地震数据去噪/卷积自编码器/深度学习/ELU激活函数Key words
seismic data denoising/convolutional auto-encoder/deep learning/ELU activation function分类
天文与地球科学引用本文复制引用
何承峻,张华..基于卷积自编码器的地震数据随机噪声压制[J].物探化探计算技术,2025,47(3):427-435,9.基金项目
江西省自然科学基金(20232BAB213077) （20232BAB213077）
赣鄱俊才支持计划-主要学科学术和技术带头人培养项目(20243BCE51012) （20243BCE51012）
