Constructing and applying a neural-network-based architectural landscape evaluation model

With the continuous improvement of living standards, people go outdoors and spend more and more time in scenic spots. The landscape architecture design that serves people in urban scenic spots attracts more and more public attention, which puts forward higher requirements for landscape architecture design that serves people in scenic spots. How to integrate better the design of all kinds of landscape architecture into nature, to serve the public better, is an urgent problem to be solved at this stage. This paper selects the evaluation indexes of urban architectural landscapes, uses the analytic hierarchy process to determine the weight of each index and quantifies six evaluation indexes to build an evaluation model of architectural landscape design. In terms of the improvement of the you only look once version 4 (YOLOv4) model, MobileNetV3 was selected as the backbone feature extraction network, and the convolution in the feature enhancement extraction network was replaced by the depth-separable volume, and an architectural landscape recognition system based on the improved YOLOv4 model was constructed. In terms of algorithm performance verification, the improved algorithm was compared with the single-shot detector, MobileNetV3, ShuffleNetV2, YOLOv3, YOLOv4 and YOLOv5s algorithms under multiple evaluation indexes. The experimental results show that the size of the model is 51.4 MB, which does not cause a large burden. The mean average precision value of the improved YOLOv4 algorithm is 93.5%, and the frames per second (FPS) value is 30 FPS, which has higher recognition accuracy and detection speed and has obvious advantages.