Review of vision-based AI approaches for infrastructure inspection using Uncrewed Aircraft Systems (UAS)

Artificial intelligence, particularly modern machine learning and deep learning, is opening up new opportunities for automating complex tasks and enhancing efficiency across various fields. One area benefiting greatly is infrastructure inspection, where AI-enabled Uncrewed Aircraft/Aerial Systems (UAS) with high resolution sensors provide a safer, contact-free alternative to traditional manual methods. This shift reduces risk and supports more effective monitoring of large or complex infrastructures. This review paper examines recent advancements in infrastructure inspection, with emphasis on how UAS collect visual data and how advanced AI models convert that data into actionable insights. It outlines key applications, evaluates the strengths and limitations of the models, and includes a critical analysis of publicly available datasets used to train vision-based inspection models. The aim is to provide a clear picture of how UAS and AI are transforming inspection practices and advancing more automated, reliable, and efficient assessment approaches.