长江大学学报（自然科学版）2025，Vol.22Issue(2)：37-44,60,9.
基于深度学习的页岩聚焦离子束扫描电镜图像多物相分割方法研究
Multi-phase FIB-SEM image segmentation of shale based on deep learning
摘要
Abstract
The pore type and composition structure of shale largely determine the reservoir performance of shale reservoirs.The pore size distribution characteristics of shale reservoirs can be well characterized by observation description and physical measurement,and focused ion beam scanning electron microscopy(FIB-SEM)imaging is the representative method to characterize the nano-scale microscopic characteristics.However,due to the influence of SEM image quality and the overlap of target gray values,the traditional threshold segmentation method cannot accurately partition the image,which makes it difficult to carry out accurate quantitative analysis of structural parameters.In this paper,FIB-SEM data of Yanchang group 73 subsegment shale is taken as the object,U-Net network architecture based on deep learning and semantic segmentation of pixel level are carried out,and multi-phase segmentation of shale pores,inorganic minerals,pyrite and organic matter is realized.Through model evaluation,the average Accuracy is up to 99.6％.The average Dice coefficient was 94.0％,which was verified by experimental data in the same area,and the effect of multi-object phase segmentation was also good.This method has laid a good rock physics foundation for the characterization of shale microstructure and multi-physical field numerical simulation.关键词
延长组页岩/聚焦离子束扫描电镜/深度学习/智能分割/U-NetKey words
Yanchang shale/FIB-SEM/deep learning/automated segmentation/U-Net分类
石油、天然气工程引用本文复制引用
夏阳,方朝强,肖占山,赵建斌,宋戴雷..基于深度学习的页岩聚焦离子束扫描电镜图像多物相分割方法研究[J].长江大学学报（自然科学版）,2025,22(2):37-44,60,9.基金项目
中国石油天然气集团公司科学研究与技术开发项目"测井数字岩石技术研究"(2021DJ4003). （2021DJ4003）
