Abstract: In order to solve the problems of semantic-based underground mine entity modeling method, such as the lack of detail level division, the small degree of freedom of model configuration, and the lack of multi-scale fine expression, parametric modeling and semantic multi-scale ideas are applied to the construction of mine multi-level of detail (LOD) model, and a mine underground space modeling method based on semantic multi-scale is proposed. It is determined that the modeling object is the entity element in mine underground space. The modeling objects are divided into roadway model and in-roadway model according to the position of the entity element. The roadway model comprises a roadway main body model and other artificial structure models, and the in-roadway model comprises a mine facility model and an equipment model. The different semantic classes are defined according to the semantic information of the entity elements. The semantic classes are divided into families, and then the families are decomposed into component elements according to their functions. Each semantic class is described by quantifiable indicators such as semantic attributes, geometric attributes, appearance attributes and characteristic attributes. The mine LOD model is designed, which includes six levels of discrete LOD models, namely roadway network model, roadway rough model, roadway fine model, main facility model in roadway, main equipment model in roadway, and other equipment models in roadway. Combining the discrete LOD model with various attributes can clearly show the level of detail and the primary and secondary relationships of the mine LOD model. The mine underground space modeling method based on semantic multi-scale can provide entity element models with different detail richness, reduce the computational complexity of scene rendering and improve the efficiency and flexibility of model construction.