A deep neural network-based green space design optimization framework for smart cities

Rapid urbanization in China has resulted in a decline in green space due to increasing population density, infrastructure expansion, and the limited integration of ecological considerations into urban planning, negatively impacting environmental sustainability and urban livability. Traditional urban greening methods are limited in scalability, responsiveness, and data integration, making it challenging to design context-aware and regulation-compliant green spaces efficiently. Hence, the research proposed a Deep Neural Network-based Green Space Design Optimization Framework (DNN-GSOF) for optimizing urban green space layouts that are ecologically efficient in smart cities. The framework utilizes a modified U-Net Convolutional Neural Network (CNN) that processes satellite imagery, environmental sensor data, and infrastructure maps to identify optimal greening zones. Secondly, it is built on a constrained deep convolutional generative adversarial network that generates diverse and regulation-compliant layout proposals. The performance evaluation module, a hybrid DenseNet-based multilayer perceptron model, quantitatively evaluates each layout across ecological, social, and economic dimensions. Compared to other models, the DNN-GSOF model has a 13.2% better layout overlap accuracy, a 17.2% higher compliance rate, and a 44.8% better FID score. The framework accelerates inference by 29.2% and reduces the mean absolute error by 40.95% for zoning compliance and 43.14% for the green space ratio. These results demonstrate that the model can generate accurate, regulation-compliant green layouts more efficiently. The proposed DNN-GSOF offers practical applications for urban planners and policymakers aiming to enhance ecological sustainability and livability in rapidly developing Chinese cities.