Integration of Small Unmanned Aircraft Systems and Deep Learning for Efficient Airfield Pavement Crack Detection and Assessment

Abstract

Airfield pavement inspection and maintenance are critical aspects of aviation infrastructure, representing a substantial portion of life-cycle costs. Longitudinal, transverse, and diagonal (LTD) cracks; corner breaks; shattered slabs in Portland cement concrete (PCC) pavement; and longitudinal and transverse (L&T) cracks of asphalt concrete (AC) pavement consist of most of the airfield pavement distresses. Traditional airfield pavement inspection methods are manual, time-consuming, laborious, and reliant on the inspector’s experience, leading to increased expenses and safety risks. This research explores the potential to automatically identify those distresses in red-green-blue images using four variants of deep learning (DL) model YOLOv8, ranging from nano to large. YOLOv8 is a widely used off-the-shelf DL object detection model that allows rapid training and easy execution. A DL training dataset of 5,273 small uncrewed aircraft systems (sUAS) collected images was developed. The transfer learning technique was used, and the dataset passed through each model 100 times for adequate training. The model exhibits mean average precision values exceeding 0.65, with varying processing times. Such accuracy showed that crack-related distress detection using DL models could enhance airfield pavement inspection efficiency.