A Hybrid CNN-SVM Classifier for Automated Pavement Performance Monitoring

This research introduces a new hybrid CNN-SVM classifier for autonomous pavement performance monitoring, combining Convolutional Neural Networks (CNNs) for deep feature extraction with Support Vector Machines (SVMs) for accurate classification. In contrast to isolated models, this hybrid model attains better computational efficiency and real-time adaptability, obviating the constraints of time-consuming manual inspections. Measured using the San Francisco Streets dataset, it achieved a stunning 97.8% accuracy, surpassing conventional approaches through optimised feature extraction and classification. Further testing on heterogeneous datasets such as Google images and the Pune-Mumbai expressway confirmed its performance across different environmental conditions. The scalable, fast solution minimises human effort, maximises realtime monitoring, and raises the bar for pavement management with its dependence on large training data. Major Findings: The hybrid CNN-SVM classifier achieved 97.8% training and 96.4% testing accuracy, outperforming standalone models in pavement defect detection. With optimal SVM settings (gamma = 0.1, degree = 3), it ensures robust, scalable performance across diverse conditions. This approach enhances real-time monitoring and maintenance strategies for road infrastructure.