A Transfer Learning-Based System of Pothole Detection in Roads through Deep Convolutional Neural Networks
Jhon Michael C. Manalo, Alvin Sarraga Alon, Yolanda D. Austria, Niño E. Merencilla, Maribel A. Misola, Ricky C. Sandil
Abstract
Pothole detection is critical in defining optimal road management solutions and maintenance. The researcher used deep learning and yolov3 to create a pothole detection system in this study. A deep learning algorithm called YOLOv3 is used to develop a model that can successfully identify potholes. The detection model had an average precision of 95.43%, and identified potholes had accuracies ranging from 33% to 69%, which is to be anticipated given the numerous various forms and sizes of potholes.
Topics & Concepts
Pothole (geology)Deep learningTransfer of learningConvolutional neural networkComputer scienceArtificial intelligenceObject detectionRangingMachine learningRemote sensingPattern recognition (psychology)GeologyTelecommunicationsPetrologyInfrastructure Maintenance and MonitoringAsphalt Pavement Performance EvaluationUnderground infrastructure and sustainability