Innovative Approaches to Road Surface Type Classification A Synergy of Vibration Data, Machine Learning, and MCNN-Based Pothole Detection

The existing road monitoring systems, which rely on manual inspections and simple sensors, are labor-intensive, unreliable, and unable of processing complicated multi-dimensional data. Consequently, it usually takes time to reveal that there is an issue with the road and to solve it, which may be dangerous if there is a threat to road users. It is suggested in the paper to integrate the vehicle-based sensors of vibration, machine learning, and the Multi-Channel Convolutional Neural Network in order to classify road surfaces and detect potholes in real time. The system can process the vibration data obtained with the help of feature extraction procedures, reducing the dimensionality using PCA and classification with Random Forest and the MCNN model. The method in question outperforms the previously existing systems with 92% accuracy, 89% precision, 91% recall, and, the maximal F1 score of 90%. The results of how the system works with all roads results in even a more accurate system: 94% smooth, 89% rough, and, 90% potholes. In addition, the system largely overcomes the disadvantages of the previous ones by almost entirely excluding the false positives and false misses in the detection of a road pavement defect; therefore, only the most efficient way a road maintenance barrier can be applied due to the massively high accuracy and ease of real-time processing.