With the high speed of urbanization construction, urban landscape design is getting more and more attention. Aiming at the problem of how to make the urban landscape have a better psychological healing effect, the study innovatively proposes to apply the theory of restorative environment to urban landscape design and builds an urban landscape element recognition model based on convolutional neural networks. From the results, the square space landscape with the lowest restorative effect of attention has the highest score on the openness evaluation index, which is 2.51. The proposed model has the highest average accuracy, which is 72.78%, and the overall accuracy is 90.49%, and the mIoU index is the highest, which is 63.8%, and the extracted picture elements are the most complete. The model of this study has the best recognition for buildings with an intersection and merger ratio of 90%. The memory occupation of this research model is 9178 Mb, and the number of references is 43.41 million, which achieves high efficiency, and the number of frames per second is 1.2. The results of the research demonstrate the good application effect of the proposed urban landscape design method, which can provide certain reference and support for urban landscape design and promote the health of urbanization construction.