드론 정사영상과 딥러닝 기반 AI 모델을 활용한 지붕 재질 및 위험요소 자동탐지에 관한 연구

This study proposes a methodology for automatically detecting roofing materials and hazardous elements using drone-based high-resolution orthophotos and the YOLOv8-seg deep learning AI model. Specifically, ultra-high-resolution imagery with a Ground Sampling Distance GSD (Ground Sample Distance) of 3cm and optimized model parameters were applied to enhance detection performance, achieving a detection accuracy of 90% (materials) and 77.8% (hazardous elements) based on the mAP50 criterion. These results not only complement traditional, labor-intensive safety inspection methods but also serve as foundational data for ensuring practical safety at small-to-medium-sized roofing construction sites and establishing a data-driven safety management system. This study is expected to provide new directions for the prevention and management of roof construction safety accidents.