基于可靠性估计的半监督地震相识别方法  

Semi-supervised seismic facies identification method based on reliability estimation

作　　者：李克文[1] 刘文龙[1] 李国庆 姚贤哲 蒋衡杰 LI Ke-wen;LIU Wen-long;LI Guo-qing;YAO Xian-zhe;JIANG Heng-jie(College of Computer Science and Technology,China University of Petroleum(East China),Qingdao 266580,China)

机构地区：[1]中国石油大学(华东)计算机科学与技术学院,山东青岛266580

出　　处：《计算机工程与设计》2024年第12期3583-3591,共9页Computer Engineering and Design

基　　金：国家自然科学基金重大基金项目(51991365);山东省自然科学基金项目(ZR2021MF082)。

摘　　要：针对地震相人工注释耗时、传统半监督方法易受不可靠伪标签干扰等缺点,提出一种基于可靠性估计的半监督地震相识别方法。使用可靠性估计网络过滤地震相伪标签中的不可靠区域,避免由错误监督信号引起的认知偏差,基于平均教师模型扩展多种类型的辅助解码器用于一致性正则化,进一步提高模型的泛化性和鲁棒性。在荷兰F3地震数据集上的实验结果表明,使用少量的标注样本MIOU可达90.33%,有效提升了容易分类混淆的地震相的识别性能。Aiming at the disadvantages of time-consuming manual annotation of seismic facies and the susceptibility of traditional semi-supervised methods to interference from unreliable pseudo-labels,a semi-supervised seismic facies identification method based on reliability estimation was proposed.A reliability estimation network was utilized to filter unreliable regions in pseudo-labels of seismic facies to avoid cognitive bias caused by erroneous supervision signals and extend multi-type auxiliary decoders based on the mean teacher model for consistency regularization to improve generalization and robustness of the model.Experimental results on the Netherlands F3 dataset demonstrate that using a few labeled samples,MIOU can reach 90.33%,effectively improves the identification performance of seismic facies that are easily confused by classification.

关 键 词：地震相识别 深度学习 半监督学习 语义分割 自训练 一致性正则化 可靠性估计 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]