Utilising Convolutional Neural Networks for Pavement Distress Classification and Detection

This paper examines deep learning models for accurate and efficient identification and classification of pavement distresses. In it, a variety of related studies conducted on the topic as well as the various identification and classification methods proposed, such as edge detection, machine learning classification informed by statistical feature extraction, artificial neural networks, and real-time object detection systems, are discussed. The study also examines the effect of image processing techniques such as gray-scaling, background subtraction, and image resizing on the performance and generalizability of the models. Using Convolutional Neural Networks (CNN) architectures, this paper proposes a model that correctly classifies images into five pavement distress categories, namely fatigue (or alligator), longitudinal, transverse, patches, and craters, with an accuracy rate of 90.4% and a recall rate of 90.1%. The model is contrasted to a current state-of-the-art model based on the You Only Look Once framework as well as a baseline CNN model to demonstrate the impact of the image processing and architecture building techniques discussed on performance. The findings of this paper contribute to the fields of computer vision and infrastructure monitoring by demonstrating the efficacy of convolutional neural networks (CNNs) in image detection classification and the viability of using CNN-based models to automate pavement condition monitoring.