Deep CNN and twin support vector machine based model for detecting potholes in road network

Potholes are a persistent issue in road infrastructure that can be responsible for safety risks and economic burdens. For effective road maintenance and ensuring the safety of road user, it is essential to detect potholes quickly. It is noticed that traditional methods are time-consuming and labor-intensive task to repair the potholes. Recently, several machine learning (ML) techniques have been integrated to design an automated pothole detection model. These methods can detect anomalies that indicate the presence of potholes by examining the features and patterns of the road surface. Furthermore, these methods are integrated into existing road infrastructure and maintenance workflows, which enable proactive maintenance strategies and resource optimization. Hence, is the present study aims to explore the efficacy of the deep convolutional neural network (CNN) and twin support vector machine (TSVM) methods for accurate identification of the potholes in road infrastructure. The deep CNN method is applied to extract the relevant feature for the road image dataset, while the twin parametric support vector machine (TPSVM) method is employed for accurate detection of potholes. The performance of deep CNN and TPSVM combination is evaluated using several performance measures. The results indicate that deep CNN-TPSVM method achieves better results than existing models for detecting the potholes in road infrastructure.