2D–3D Geological features collaborative identification of surrounding rock structural planes in hydraulic adit based on OC-AINet

Abstract Timely acquisition of geological structure information is crucial for large-scale water conservancy and hydropower projects. Understanding surrounding rock structure is critical for rock deformation analysis and geological evaluation of hydraulic adits. This study proposes a new method of OC-AINet to achieve efficient and accurate recognition of geological structure information of large hydraulic adits. An ordinary digital camera is used to obtain images of the Heishanxia adit, and dense point clouds are generated through multi-view stereo reconstruction. 2D–3D feature semantic super-pixel blocks are segmented by combining color, texture, and other geological semantics with AINet, and then a hierarchical regional clustering is carried out for rock structural planes extraction. The experimental results show that the difference between the rock structural plane identification error of the proposed method and that of the traditional manual method is less than 5 degrees, which meets the requirements of geological analysis. In addition, it has higher efficiency and safety than the manual method. The proposed OC-AINet has significantly improved the accuracy and reliability of the segmentation results and promotes the application of photogrammetry technology in geological scenes, especially in hydraulic adit environments.