Abstract: Accurate identification of joint orientations on the tunnel face is a critical indicator for surrounding rock stability evaluation and decision- making on excavation and support timing. To address the limitations of traditional manual contact measurements, which are low in acquisition efficiency and high in operational risk, as well as the constraints of a single digital photography or 3D laser scanning technology, a method for tunnel face joint recognition integrating laser point clouds and digital images is proposed. This method integrates the advantages of 3D structural plane orientation analysis from point clouds and deep-learning-based extraction of 2D joint traces from images, and a 3D tunnel face model with both geometric accuracy and texture information is constructed. Based on this model, automatic segmentation of structural planes and calculation of their orientation are achieved, as along with pixel-level identification of joint trace and calculation of the rock mass integrity index. Through multi-source data acquisition, joint calibration, and fusion registration, a colored point cloud model is generated. Automatic structural plane segmentation is realized using point cloud normal vector calculation and an improved region-growing algorithm. A convolutional neural network suitable for joint trace extraction is constructed, and combined with conditional random field post-processing, the identification and extraction of joint traces from tunnel face images are realized. Practical engineering applications demonstrate that this method can effectively identify the orientations of structural planes on the tunnel face surrounding rock, with the calculation errors within 4° compared to manual geological compass measurements, the joint trace identification achieves an IoU of 0.82 and an F1-score of 0.89. The rock mass integrity index can also be calculated, providing reliable data support for tunnel surrounding rock classification and blasting parameter optimization.