A Fast and Effective Machine Learning Approach for Road Cracks Classification

Road transportation is the backbone of the economic activities of any nation. Millions of commuters use national and state highways and local municipal roads for daily operations. Road maintenance is an indispensable part of the concerned authorities for hassle-free and safe transportation. Poor maintenance of the road and its potholes could cause fatal injuries to the commuters. As the length of these nationwide road networks is overwhelming, the modern-day machine learning-based approach can efficiently detect and classify such a variety of road cracks and send information to the appropriate parties. This paper introduces a straightforward machine learning-guided pipeline that can successfully classify the different types of cracks on the road. It is also compared with state-of-the-art Convolutional Neural Network (CNN) models. Our proposed machine learning approach and different feature selection techniques provide consistent and competitive results in a short training time. It is observed that Orthogonal Matching Pursuit (OMP) outperforms other used feature selectors in terms of selected feature subset size, classification performance, and training time. The final, conclusive result is as high as ∼ 97% accuracy obtained from OMP feature selection and Optuna-based tuned Random Forest Classifier.