Study on rapid construction and web loading performance optimization of massive city 3D models based on deep learning

Three-dimensional urban models have extensive applications in various fields, but efficient construction and smooth display on web platforms are critical issues. This paper proposes a method for rapid construction of massive 3D models based on deep learning, designing an end-to-end convolutional neural network that integrates remote sensing imagery and geographic information data to automatically generate large-scale building models. Meanwhile, strategies such as model optimization, data compression, progressive transmission, and GPU parallel rendering are adopted to address web loading performance issues. Experiments demonstrate that this method can efficiently construct models covering 4 million buildings across the city, meeting mapping standards, and achieve rapid loading and smooth rendering of 3D scenes on different hardware environments, providing strong support for the development in related fields.