首页|基于空间注意力机制的Mask R-CNN致密储层岩石薄片图像鉴定

基于空间注意力机制的Mask R-CNN致密储层岩石薄片图像鉴定

扫码查看
原文链接
  • 国家科技期刊平台
  • NETL
  • NSTL
  • 万方数据
针对陆相致密储层岩石薄片鉴定识别难、制片成本高、时间消耗长和人为主观强等难题,选取鄂尔多斯盆地临兴区块上古生界和松辽盆地三肇凹陷扶余油层为靶区,提出一种基于深度学习的致密油储层岩石薄片人工智能鉴定方法,引入图像预处理技术去除岩石薄片图像噪声并统一图像像素大小,构建空间几何增广机制,基于空间注意力机制改进Mask R-CNN算法,并将上述方法应用于实例靶区进行有效性验证.结果表明:图像预处理技术能够在保障图像特征的前提下,有效提高图像质量,减少噪声干扰;空间几何图像增广机制能够在在一定程度上增加可用样本的数量;基于空间注意力机制的Mask R-CNN算法可以同时完成复杂岩石薄片成分的分割与智能识别工作,分割精度在不同数据集情况下的平均精度为 89.2%,整体识别准确率为 93%,适用于致密油储层岩石薄片特征鉴定.
Image identification of rock slices of Mask R-CNN tight oil reservoir based on spatial attention mechanism
Aiming at the difficult identification,high production cost,long time consumption and strong human subjective of rock thin section identification in continental tight reservoirs,a deep learning based artificial intelligence method for thin sec-tion identification of tight oil reservoirs was proposed by selecting the Upper Paleozoic in Linxing Block of Ordos Basin and Fuyu reservoir in Sanzhao Sag of Songliao Basin as target areas.Through the introduction of image preprocessing technology to remove the noise of rock slice image and unify the size of image pixels,a spatial geometry enhancement mechanism was constructed,and the Mask R-CNN algorithm was improved based on the spatial attention mechanism.The effectiveness of the above method was verified by applying it to the sample target area.The results show that the image preprocessing technology can effectively improve image quality and reduce noise interference under the premise of guaranteeing image features.The spatial geometry image augmentation mechanism can increase the number of available samples to some extent.The Mask R-CNN algorithm based on the spatial attention mechanism can simultaneously complete the segmentation and intelligent identi-fication of complex rock sheet components.The average accuracy of segmentation accuracy in different data sets is 89.2%,and the overall identification accuracy is 93%,which is applicable to the characterization of rock sheets in tight oil reser-voirs.

tight oil reservoirrock thin sectiondeep learningMask R-CNN algorithmsegmentation and recognition

李春生、刘涛、刘宗堡、张可佳、刘芳、刘晓文、田梦晴、白玉磊、尹靖淞、卢羿州

展开 >

东北石油大学计算机与信息技术学院,黑龙江大庆 163318

东北石油大学地球科学学院,黑龙江大庆 163318

致密储层 岩石薄片 深度学习 Mask R-CNN算法 分割与识别

国家自然科学基金面上项目国家青年科学基金项目中国石油科技创新基金项目黑龙江省优秀青年科学基金项目黑龙江省自然科学基金项目黑龙江省创新型科研人才培养计划项目

42172161421021732020D-5007-0102YQ2020D001LH2020F003UNPYSCT-2020144

2024

10.3969/j.issn.1673-5005.2024.04.003

中国石油大学学报(自然科学版)
中国石油大学

中国石油大学学报(自然科学版)

CSTPCD北大核心
影响因子：1.169
ISSN：1673-5005
年,卷(期)：2024.48(4)
  • 14