Ground Landscape Urban Element Extraction Method Based on Optimized Convolutional Neural Network Technology

With the continuous advancement of modernization and urbanization, the importance of ground landscape design in urban environments is becoming increasingly prominent. To carry out reasonable design and planning of urban landscapes, this study uses a convolutional neural network to construct a DeepLab-v3+ground landscape design and planning model. The polarized selfattention mechanism is adopted to improve the accuracy of the model in extracting ground landscape elements in complex environments. Meanwhile, this study utilizes transfer learning to optimize the model and improve its efficiency. The results showed that the maximum accuracy, recall, and F1 value of the research model reached 94.02%, 91.93%, and 93.45%, respectively. The landscape recognition accuracy of the model in practical applications reached 92.3%, with a video memory usage of 8743MB and a prediction speed of 1.2FPS. In summary, the ground landscape design and planning method based on optimized convolutional neural network technology can accurately and quickly classify urban remote sensing images, which is helpful for the actual planning and design of urban ground landscapes.