Research On Multi-type 3D Terrain Generation Technology Based On Deep Learning

Terrain modeling has always been a very important topic in computer graphics.With the rapid development of games,film and television,military simulation and aircraft simulation,the demand for terrain generation is also increasing sharply.How to efficiently generate high-quality terrain to meet a large number of requirements has become a challenging problem.The new development of deep learning provides a new idea for terrain modeling that is different from traditional methods.This paper uses terrain classifiers to calculate the topographic features of different terrains,and combines the features of each terrain into an implicit vector to control the type of terrain generated by the network.After the user selects the desired terrain type,the generative adversarial network designed in this paper can transform the user-designed sketch into a desired terrain height map.In addition,users can choose transition states between multiple implicit vectors to create more meaningful terrain.This paper designs rendering scripts for textures and vegetation on various terrains in software such as Unreal Engine 4,to cover different styles of terrain with texture and vegetation,so as to obtain highly realistic large-scale terrain.