Abstract: The classification of surrounding rock is an important basis for determining and adjusting the rationality of tunnel construction schemes; therefore, the traditional surrounding rock classification method can be improved to reduce economic losses, safety accidents, and other problems caused by the difference between the actual classification of surrounding rock and geological surveying results. Based on the Dafalang tunnel of the Wenma expressway in Yunnan, threedimensional (3D) reconstruction, stitching of images, Unet neural networks, and surrounding rock characteristics of uniaxial compressive strength are considered to realize the recognition of structural plane features and the rapid evaluation of surrounding rock classification based on the integrity and strength characteristics of the rock mass. A digital camera collects images of the tunnel face and surrounding wall to create a 3D model using 3D image reconstruction. Second, the projection algorithm and image Mosaic technology are used to generate a highdefinition image of the tunnel face. Third, using Unet neural networks, a joint outline analysis is performed to automatically obtain tunnel face integrity information. Finally, the surrounding rock of the excavation face is classified using the BQ classification method, taking into account other rock characteristics. The results show that clear images of the tunnel face can be obtained using the recommended method. The field application results are more in line with the actual situation than the original design, which has a good application.

Key words:  , highway tunnel, threedimensional modeling, image stitching, Unet neural networks, joint, BQ classification