Scholarlink

<p>Deep carbonate reservoirs represent one of the most important types of hydrocarbon reservoirs, and accurate identification and characterization of their reservoir space structures remain among the key technical challenges for efficient exploration and development in the petroleum industry. To address the low accuracy and limited classification capability of conventional void recognition methods in electrical image logs of deep carbonate formations, this study proposes an integrated technical workflow that combines data preprocessing, deep-learning-based recognition, and quantitative parameter characterization. First, multiple image restoration algorithms were compared, and the radial basis function (RBF) interpolation algorithm was selected to eliminate blank stripes in electrical image logs. A joint annotation method was established to achieve precise multi-type labeling of voids. By integrating dynamic and static imaging data, a dual-channel dataset was constructed to effectively fuse complementary dynamic and static image information. Subsequently, an adaptive dynamic fusion module (EC-Gate) was designed and embedded into cross-layer connection structures. A deep-learning semantic segmentation model, GateNet, was developed for multi-type cavity recognition. Comparative experiments demonstrated that GateNet exhibited good generalization performance within the study area and high recognition accuracy, outperforming multiple classical models in both recognition precision and completeness, achieving the highest evaluation metrics (mean intersection over union [MIoU] = 90.89%, pixel accuracy [PA] = 94.24%). Finally, an integrated workflow of “intelligent recognition – morphological optimization – parameter characterization” was established and successfully applied to cavity identification and areal porosity extraction in a blind well section. This study provides a novel technical approach for refined void characterization in carbonate reservoirs and may offer potential application value for oil and gas exploration and development.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep learning and multi-view image fusion for identification of karst voids and areal porosity quantification in deep carbonate reservoirs: a case study of the Lungu-7 block, Tarim Oilfield, China

  • 期刊论文

  • 发表日期：

    2026年7月17日

查看全文
  • Zhuolin Li,
  • Guoyin Zhang,
  • Jianli Lin,
  • Xin Zhang,
  • Jinqiang Tian

摘要

Deep carbonate reservoirs represent one of the most important types of hydrocarbon reservoirs, and accurate identification and characterization of their reservoir space structures remain among the key technical challenges for efficient exploration and development in the petroleum industry. To address the low accuracy and limited classification capability of conventional void recognition methods in electrical image logs of deep carbonate formations, this study proposes an integrated technical workflow that combines data preprocessing, deep-learning-based recognition, and quantitative parameter characterization. First, multiple image restoration algorithms were compared, and the radial basis function (RBF) interpolation algorithm was selected to eliminate blank stripes in electrical image logs. A joint annotation method was established to achieve precise multi-type labeling of voids. By integrating dynamic and static imaging data, a dual-channel dataset was constructed to effectively fuse complementary dynamic and static image information. Subsequently, an adaptive dynamic fusion module (EC-Gate) was designed and embedded into cross-layer connection structures. A deep-learning semantic segmentation model, GateNet, was developed for multi-type cavity recognition. Comparative experiments demonstrated that GateNet exhibited good generalization performance within the study area and high recognition accuracy, outperforming multiple classical models in both recognition precision and completeness, achieving the highest evaluation metrics (mean intersection over union [MIoU] = 90.89%, pixel accuracy [PA] = 94.24%). Finally, an integrated workflow of “intelligent recognition – morphological optimization – parameter characterization” was established and successfully applied to cavity identification and areal porosity extraction in a blind well section. This study provides a novel technical approach for refined void characterization in carbonate reservoirs and may offer potential application value for oil and gas exploration and development.