A Hybrid Framework for Pavement Crack Detection Using Deep Feature Extraction and Machine Learning for Smart Road Monitoring

Early automated pavement crack detection and classification is an essential task to sustain the life of the road and ensure public safety in the context of a smart city. Therefore, the aim of this study is to use hybrid deep learning to develop a system for detecting pavement cracks. In this paper, the proposed system operates by first processing road surface crack images through three deep learning models, namely Resnet50, InceptionV3 and Xception for feature extraction, followed by five machine learning models, namely Random Forest (RF), Adaptive Boosting (AdaBoost), K-Nearest Neighbors (KNN), Decision Tree (DT), and Naive Bayes (NB). For accurate pavement crack classification as cracked and non-cracked in roads. The key objective of this approach to develop an intelligent pavement cracks prediction system for smart road applications. The subsequent system provides a model for pavement crack detection and classification. As well, this paper introduces an efficient framework based on hybrid deep learning approaches for feature extraction from pavement cracks and classify them using the machine learning classifiers. The proposed framework demonstrated best performance with the RF classifier, achieving an accuracy of 83% using both InceptionV3 and Xception deep features. In comparison, the other models such as AdaBoost (78–79%), KNN (58–69%), DT (70–71%), and NB (61–72%) exhibited comparatively lower accuracy. The results show the robustness and consistency of the RF model with all metrics. Accordingly, the proposed approach can effectively performing pavement crack detection for real-time monitoring in smart transportation systems.