SM-YOLO: A Novel Small Components Detection Algorithm for Power Delivery Infrastructure

Power line components detection is an important and challenging task in intelligent transportation system and smart grid operation. However, the contradiction between huge number of maintenance components and few number of workers makes unmanned aerial vehicles based untouchable inspection strategy become increasingly popular. In this paper, we proposed a novel YOLO based visual inspection method, which can improve small object detection problem. To use YOLO-v5 as backbone, then a core query mechanism constructs a high-resolution feature-pyramid based object detector. To verify the performance of proposed method, we carry on experiment based on a novel self-collected dataset. It is demonstrated experimentally that this method can accurately locate small object on UAV based real-time image data.