Abstract: To solve the problem of difficult restoration of Dunhuang murals, especially for the damaged surface murals on the top of caves, a Dunhuang mural restoration algorithm (HASM) based on spatial structure prediction combined with line drawing assisted guidance is proposed. To highlight the spatial structure of the mural, the spatial structure of the mural and the surface information of the cave were considered for the first time. A wireframe parser was used, and a simplified 4D geometric vector space was introduced to encode the line segments in the wireframe. The line segments were extracted and connected using connection generation, matching, and verification algorithms. The line segment masking algorithm was used, combined with the spatial structure prediction module and mural edge extraction, and guided by the complete mural line drawing dataset. A spatial box assisted parsing network (L-HAT) for mural images was designed. To reduce the loss of mural edge information during the restoration process and bring in spatial structure attention, this paper designs a mural restoration model. This allows the model to not only repair flat damages, also maintain a good repair effect on murals with large damaged areas. The experimental results show that this model effectively preserves the structural information of the cave during the restoration process, restores the surface structure information of the cave, and solves the problem of difficult restoration of cave murals on curved surfaces.