西安科技大学学报2025，Vol.45Issue(3)：481-490,10.DOI:10.13800/j.cnki.xakjdxxb.2025.0306
基于DeepLabv3+网络的煤体孔隙识别及分析
Coal pore identification and analysis based on DeepLabv3+network
摘要
Abstract
The pore structure of coal matrix affects the occurrence and migration of gas,which is crucial for gas emission prediction and safe coalbed methane mining.To achieve accurate characterization of coal pores,two coal samples from Xiaobaodang(XBD)and Sangshuping(SSP)mining areas were taken as examples.The pore distribution images of coal were obtained through focused ion beam scanning e-lectron microscopy(FIB-SEM)experiments,with a dataset established.A DeepLabv3+model for coal pore image recognition and segmentation was constructed based on machine learning methods,and com-parative experiments were conducted with classical network models PSPnet and UNet to achieve rapid recognition and analysis of coal pore structure.The results show that the DeepLabv3+network has good segmentation performance,with an average intersection ratio of 92.71％,which is 12.67％and 2.32％higher than that by PSPnet and UNet networks,respectively.It has strong recognition ability for micro and nano pores.The XBD coal sample has a higher distribution of large pores with a pore size greater than 50 nm,accounting for 55.02％of the total pores.It mainly consists of angular gravel pores,interg-ranular pores,and dissolution pores,with good pore connectivity.The number of transition pores and macropores with a pore size of 2～50 nm in SSP is relatively large,accounting for76.04％of the total pores.The morphology is relatively simple,and the average roundness reaches 0.531 μm.However,the poor connectivity between pores is not conducive to the migration of gas,which is consistent with the measurement results of gas emission.The research results confirm the good applicability of the Deep-Labv3+model in coal pore image segmentation,providing a reference for the characterization and anal-ysis of coal pore structure.关键词
孔隙结构/瓦斯涌出量/聚焦离子束扫描电镜/机器学习/DeepLabv3+模型Key words
pore structure/gas emission rate/FIB-SEM/machine learning/DeepLabv3+model分类
矿山工程引用本文复制引用
刘纪坤,张博浩,王翠霞,赵兰华,徐栋梁..基于DeepLabv3+网络的煤体孔隙识别及分析[J].西安科技大学学报,2025,45(3):481-490,10.基金项目
国家自然科学基金项目(52204235) （52204235）
