Asphalt pavement cracking classification using convolutional neural networks

Abstract

Asphalt pavements on roads are subject to cracking due mainly to overload and weather over time. Detecting the presence of cracking is an essential part of road maintenance systems. Traditional road defect detection methods are time-consuming, dangerous, labor-intensive, and subjective. An alternative is to use digital images of roads, collected by cars or unmanned aerial vehicles, to feed automated asphalt damage identification models. Thus, automated defect detection systems could quantify the quality of road surfaces and help prioritize and plan road network maintenance, thereby preserving and extending the useful life of roads. This work aims to investigate the use of machine learning for detecting and classifying cracking in road pavements using convolutional neural networks. Transfer learning and data augmentation techniques were used to build classification models. The results indicated that the model was able to identify cracking in the images of roads with an accuracy of 99%. When using the model to classify the cracking subtypes, the model presented an overall accuracy of 78%.