摘要
储层裂缝研究一直是油气勘探开发中的一个重点,其中最直观的方法就是对露头岩石裂缝进行观察。由于人工绘制效率低,主观性强,所以露头岩石裂缝识别一直是个难点。针对该难点,利用无人机倾斜摄影技术大幅提高测绘效率,同时提出一种基于改进的DeepLabV3+模型,实现露头岩石裂缝的精准分割。首先,选用MoblieNetV2作为主干网络,降低模型的参数量;其次使用跨层链接的ASPP(atrous spatial pyramid pooling),有效地提取和融合特征,同时引入SP(strip pooling)条形池化分支,增强特征图的语义信息和空间分辨率;最后加入SE(squeeze and excitation)注意力模块,增强对裂缝关键边缘与密集分布位置特征的提取能力。实验结果显示,改进的DeepLabV3+模型岩石裂缝均值平均精度达到83.77%,相比改进前DeepLabV3+模型的71.24%提升了12.53个百分点,平均交并比由68.21%提升到78.84%,准确率由74.83%提升到91.66%,参数量由5.81 M增加到15.14 M,具备较好的移动部署性能。改进的DeepLabV3+模型在露头岩石裂缝分割任务中表现出优越的分割性能。
Reservoir fracture research has always been a focal point in oil and gas exploration and development,with the most direct method being the observation of rock fractures in outcrop areas.Due to the low efficiency and high subjectivity of manual delineation,the identification of rock fractures in outcrop areas has long been a challenging issue.To address this challenge,this paper utilizes drone oblique photography technology to significantly enhance mapping efficiency and proposes an improved DeepLabV3+semantic segmentation model for the precise segmentation of rock fractures in outcrop areas.Initially,the MobileNetV2 is selected as the backbone network to reduce the model’s parameter count.Subsequently,ASPP(atrous spatial pyramid pooling)with cross-layer connections is employed to effectively extract and integrate features.Additionally,SP(strip pooling)is introduced to enhance the semantic information and spatial resolution of the feature maps.Finally,a SE(squeeze and excitation)attention module is incorporated to bolster the model’s ability to extract features from critical edges and densely distributed areas of fractures.Experimental results demonstrate that the improved DeepLabV3+model achieves an mean average precision of 83.77%for rock fractures,which are 12.53 percentage points increase from the original DeepLabV3+model’s 71.24%.The average Intersection over Union has been improved from 68.21%to 78.84%,and the accuracy has been enhanced from 74.83%to 91.66%.Although the model’s parameter count has increased from 5.81 M to 15.14 M,it still possesses good performance for mobile deployment.The experimental results indicate that the improved DeepLabV3+model presented in this paper exhibits superior segmentation performance in the task of rock fracture segmentation in outcrop areas.
作者
白凯
张瑶崴
王举
印森林
BAI Kai;ZHANG Yaowei;WANG Ju;YIN Senlin(Hubei Key Laboratory of Drilling and Production Engineering for Oil and Gas(Yangtze University),Wuhan 430100,Hubei;School of Computer Science,Yangtze University,Jingzhou 434023,Hubei;Research Institute of Mud Logging Technology and Engineering,Yangtze University,Jingzhou 434023,Hubei)
出处
《长江大学学报(自然科学版)》
2025年第5期66-75,共10页
Journal of Yangtze University(Natural Science Edition)
基金
中国石化油气成藏重点实验室开放基金项目“基于无人机倾斜摄影的地质剖面全景可视化”(zshkfjj2021-wxs01)
油气钻采工程湖北省重点实验室开放基金项目“基于深度学习的页岩气钻井漏失事故风险智能预测方法研究”(YQZC202511)。
关键词
无人机
露头岩石裂缝
倾斜摄影
裂缝识别
深度学习
unmanned aerial vehicle
outcrop rock cracks
drone oblique photography
crack identification
deep learning
