Abstract: Lithology is a critical parameter in rock engineering design and construction,significantly influencing the stability and safety of tunnels.In order to solve the existing problem of insufficient identification accuracy of mixed lithology at the tunnel face,this paper fuses and analyzes the image data of rock mass scale and tunnel face scale,and proposes an intelligent identification method of mixed lithology probability distribution at the tunnel face based on two-scale machine learning.First,the ResNet-101 image classification algorithm is used to classify and identify lithology at the rock block scale,based on visual features such as color,texture,and mineral composition of the rock blocks produced by tunnel blasting.Then,the UNet++ semantic segmentation algorithm is applied to segment and identify the lithology probability distribution at the tunnel face scale.Finally,the identification results at the rock block scale are intelligently matched with the lithology probability distribution at the tunnel face scale,achieving a highly accurate prediction of lithology distribution at the tunnel face.Rock block and tunnel face datasets were constructed for training and testing.The results show that compared with the direct use of semantic segmentation algorithm to identify the mixed lithology probability distribution of tunnel face,the proposed method effectively avoids misidentification of non-target lithologies of the face,and significantly improves the lithology identification accuracy.