Abstract: To address the limitations of traditional artificial sketching, an extraction method for the surrounding rock structural surface features of the tunnel face based on a threedimensional(3D) geometric image is proposed. First, the surrounding rock data of the tunnel face is collected using a camera and LiDAR, enabling the creation of a 3D realistic model of the tunnel face. Then, the structural surfaces of the surrounding rocks are identified by analyzing the difference in their normal vectors within the  3D model. Furthermore, a convolutional neural network algorithm is used to train software in recognizing the surrounding rock joints based on geological engineers′ sketches. Subsequently, the recognition results of the 3D model and the image recognition training are combined with the recognition to achieve automatic recognition of the surrounding rock joints and tracks of the tunnel face. Finally, the integrity information of the surrounding rock at the tunnel face is calculated based on the joint number of rock volumn〖WT5BZ〗 Jv 〖WT５《TNR》〗of tunnel face′s surrounding rock. The method has been applied to the tunnel faces of Huangkeng and Lingshang village tunnels of the ChizhouHuangshan highspeed railway to analyze the integrity of the surrounding rock of the tunnel face. Results demonstrate that this method can obtain a highprecision 3D model of the tunnel face and quickly and accurately assess the integrity of the surrounding rock, facilitating the compilation of geological information about the tunnel.

Key words:  , railway tunnel, tunnel face, threedimensional reconstruction, convolutional neural network, rock joint, rock mass completeness