Abstract: With the rapid development of virtual reality technology,the modeling of three-dimensional virtual scenes in mines has shown broad application prospects in areas such as mine safety management,training,and disaster prevention. However,traditional three-dimensional modeling methods usually rely on manual operations,which are time-consuming and have limited accuracy. Therefore,a method for modeling three-dimensional virtual scenes in mines based on an improved generative adversarial network is proposed. Firstly,three-dimensional laser scanning technology is used to obtain the actual terrain data of the mine and construct a high-precision point cloud model. Then,through preprocessing and feature extraction,a conditional generative adversarial network is designed to improve the stability and realism of the generated model. The point cloud data is input into the improved generative adversarial network. The generative network generates realistic three-dimensional virtual scenes of the mine by learning the spatial distribution features of the point cloud data;the discriminative network is used to evaluate the authenticity of the generated scenes and guide the optimization of the generative network. The experimental results show that the improved generative adversarial network method achieves a detail retention rate of 95. 8%,and it only takes 15 s to generate a three-dimensional model. It is superior to traditional methods in terms of detail retention,authenticity,and modeling efficiency,providing strong support for the modeling of three-dimensional virtual scenes in mines.

Key words: mine 3D modeling,3D laser scanning,virtual reality,generative adversarial network