基于混合神经网络的地震断层自动识别方法  

Automatic seismic fault recognition method based on hybrid neural network

作　　者：魏彤 尹成[1] 丁峰[1] Wei Tong;Yin Cheng;Ding Feng(School of Earth Science and Technology,Southwest Petroleum University,Chengdu 610500)

机构地区：[1]西南石油大学地球科学与技术学院,四川成都610500

出　　处：《石化技术》2022年第4期123-124,共2页Petrochemical Industry Technology

摘　　要：本文提出的断层自动识别方法HCA能够有效改善这一现状。HCA利用混合卷积进行非连续特征提取,然后使用通道注意力机制捕捉断层证据,最后通过ASPP模块进一步提取地震图像的多尺度特征用于断层结果输出。在设计实验中本模型的准确率为0.9502,高于其他模型,能够用于实际地震解释节省时间和人力成本。The automatic fault identification method HCA proposed in this paper can effectively improve this situation.HCA uses hybrid convolution for discontinuous feature extraction,then uses the channel attention mechanism to capture fault evidence,and finally the multi-scale features of seismic images are further extracted by the ASPP module for fault result output. The accuracy of this model in the designed experiments is 0.9502,which is higher than other models and can be used for actual seismic interpretation to save time and labor costs.

关 键 词：断层识别 深度学习 多尺度卷积 

分 类 号：P631.4[天文地球—地质矿产勘探]