Artistic collaborative design optimization of urban public architecture based on faster region convolutional neural network and artificial intelligence

This study proposes a collaborative design optimization framework based on the improved Faster Region-based Convolutional Neural Network (Faster R-CNN). It aims to handle the current situation of over-reliance on subjective experience and low collaborative efficiency in the art design of urban public architecture. Through domain adaptation of the standard Faster R-CNN model, modules such as architectural image pre-training, bidirectional multi-scale feature fusion, texture enhancement, and style-aware attention are introduced; these can remarkably improve the model's detection accuracy for complex artistic elements. On this basis, a human-machine collaborative system integrating intelligent analysis, visual interaction, and feedback learning is constructed. This approach combines the quantitative analysis capability of artificial intelligence (AI) with the professional creativity of designers. Experiments are conducted based on the public ADE20K dataset. The improved model achieves a mean average precision (mAP) of 73.5%, representing an 8.4% increase compared with the baseline model. In actual collaborative scenarios, the proposed system can effectively improve the style consistency and specification compliance rate of design schemes while shortening the average design cycle. This study provides systematic theoretical methods and practical paths for AI to empower architectural art design, promoting the transformation of the design process toward a data-driven, human-machine collaborative, and intelligent direction.