随着地震勘探地区的复杂化和地震装备及采集技术的发展,采集到的中低信噪比地震数据数量急剧增加,传统的初至拾取方法由于效率低、精度差己不能满足资料处理需求.针对这一问题,在充分分析传统语义分割网络方法基础上,提出了一种由编码器和解码器两部分组成的端到端的深度学习网络模型融合自注意力机制的空洞卷积空间金字塔池化(Adaptive Aggregation Net,AANet).实验结果表明,训练后的网络模型能够高质量地拾取中低信噪比地震数据中的初至时刻,在测试集中的预测评价指标均交并比(Mean Intersection over Union,MIoU)达到99.9%.AANet提高了中低信噪比地震数据初至拾取的精度和效率,有良好应用前景.
Application research of deep learning in first-arrival pickup for seismic data with low to medium signal-to-noise ratio
As seismic exploration regions become increasingly complex and with the advancement of seismic equipment and acquisition technologies,there has been a sharp rise in the quantity of seismic data with low to medium signal-to-noise ratios(SNR).Traditional first-arrival picking methods,characterized by low efficiency and accuracy,can no longer meet the demands of data processing.Addressing this issue,this paper proposes an end-to-end deep learning network model(AANet),which is composed of an encoder and a decoder,building upon a thorough analysis of conventional semantic segmentation networks.Experimental results show that the trained network model is capable of high-quality first-arrival picking for seismic data with low to medium SNR,achieving a predicted MIoU(Mean Intersection over Union)of 99.9%on the test set.AANet improves the accuracy and efficiency of first-arrival picking of seismic data with low to medium SNR,offering promising prospects for industrial application.
国家科技期刊平台
NSTL
万方数据
