Detecting Buildings from Remote Sensing Imagery: Unleashing the Power of YOLOv5 and YOLOv8

Abstract

Building detection plays a crucial role in various applications, for example, urban planning, infrastructure development, and disaster assessment. In the past few years, single-stage object detection algorithms, specifically the YOLO (You Only Look Once) models, have gained significant attention due to their real-time performance and accuracy. This research work concentrates on the application of YOLOv5 (nano (n), small (s), medium (m), and large (l)) and YOLOv8 (n, s, and m) for building detection from remotely sensed images. For experimentation, the popular SZTAKI-INRIA building detection benchmark dataset is used. Further the standard indicators namely f-score, recall, precision and mean average precision (mAP) are computed for quantitative analysis. The results showed that the performance of YOLOv5l (f-score=0.915) is best followed by YOLOv8m with an f-score of 0.905. The results exhibit the potential of both models in accurately detecting buildings from remote sensing data.