An intelligent hybrid YOLO–CNN–LSTM framework for real-time road infrastructure monitoring and analysis

Deep learning has shown great potential for intelligent road infrastructure management, yet accurate and scalable assessment remains challenging due to the limitations of manual inspection and single-modality approaches. To address these issues, this study presents a hybrid framework that integrates multimodal IoT sensing, Digital Twin (DT) modeling, and a hybrid YOLO, CNN, BiLSTM architecture for real-time pavement monitoring and severity prediction. The framework leverages heterogeneous sensor data and DT simulations to provide context-aware anomaly detection, focusing on cracks, potholes, faded markings, and snow-covered surfaces. By combining spatial and temporal feature learning, the system enables both immediate detection and predictive assessment of deterioration trends. Experimental validation confirms its superior performance over state-of-the-art models, achieving high temporal efficiency (11.02 ms), robust detection accuracy (Precision = 96.26%, Sensitivity = 95.52%, Specificity = 95.73%, F-Measure = 96.25%), and low prediction error (0.21 ±0.01). Overall, the proposed framework offers a reliable, scalable, and proactive solution for intelligent road infrastructure monitoring and maintenance.