Automating Pavement Health Assessment: Leveraging Machine Learning for Efficient Pavement Crack Detection and Classification

Abstract

Pavement degradation, including cracks, are a major road maintenance issue and a hazard to the safety of infrastructure. Conventional manual checking systems are time consuming and are subject to human error making automated cracking systems very essential. This research discusses four machine learning (ML) classifiers, including k-Nearest Neighbors (kNN), Support Vector Machine (SVM), Neural Network (NN) and Logistic Regression (LR), as efficient to automate the process of pavement crack detection based on image classification. The models were trained and tested on a set of 360 images (180 cracked and 180 uncracked pavement). The findings revealed that SVM, NN and LR had 100 percent accuracy in the classification of cracked and uncracked pavement whereas the kNN had 97.2 percent accuracy. Although the performance decreased slightly, kNN was found to be very reliable thus it can be effectively used to carry out the task of detecting cracks. Measures of evaluation, such as Area Under Curve (AUC), Classification Accuracy (CA), F1 Score, Precision, Recall, and Matthews Correlation Coefficient (MCC) showed that all models performed excellent. The results indicate the promise of ML in the domain of pavement condition assessment automation, which will result in the improved and less expensive infrastructure management. Future research must aim at extending the models to other more complicated situations, such as multi-class classification and real-time monitoring.