In order to address the problems of long time cycles and complex processing work in the current use of 2D data for 3D modeling,a 3D virtual landscape system was constructed based on the analysis of the application characteristics of unmanned aerial vehicle tilt photogrammetry technology.Subsequently,an improved Convolutional Neural Networks(CNN)structure was designed on the TensorFlow platform and applied to a 3D virtual landscape system.The results show that in the self-made 3D landscape dataset,the improved CNN has a high accuracy rate of 96.36％for feature extraction of green landscapes.In the visual effect and functional verification,the proposed 3D virtual landscape system can score up to 97.5 points,and the shortest time is only 108s.It shows that the designed 3D virtual landscape system can effectively extract the features of various landscape types,and has better visualization effect and functionality,which provides a method reference for the construction of real 3D model and smart city.
维普
万方数据