Artificial intelligence-powered graph neural network-YOLO framework for real-time detection of environmental hazards in sustainable cities

The current detection systems have limited real-time potential, have poor spatial relationship modeling, and accuracy of multi-hazard recognition, and hence the current challenges to sustainable development of cities are unprecedented due to the environmental degradation in cities. This study presents a new hybrid architecture of the application of Graph Neural Networks (GNN) and You Only Look Once version 8 (YOLOv8) to solve the issue of wholesome real-time identification of environmental hazards in smart cities. The proposed GNN-YOLOv8 model uses spatial-temporal graph convolutional layers to model the environmental sensor-networks and establishes the benefits of the developed visual hazard detection using advanced object detection abilities. The pipeline used includes three stages, which are constructing a graph based on heterogeneous sensor data, extracting features with the help of attention-based GNN modules, and classifying hazards with changed YOLOv8 and adaptable anchor mechanisms. The one-way ANOVA statistical analysis indicated that the results showed their significance in the difference of performance between hazard categories. The framework delivered real time processing rates of 67.3 frames per second and a latency of 14.8 milliseconds and could therefore be deployed to constrained resource edge computing devices. The study provides a scalable interpretable framework of sustainable city environmental surveillance that has an impressive contribution to the development of smart city infrastructure and climate change management meta compliances.