YOLOv8-CVIFB: A Fast Object Detection Algorithm for UAV Power Patrol Based on YOLOv8 Heterogeneous Image Fusion

Visual inspection of power equipment has been a hot topic in recent years. In this field, people either focus on object detection, image fusion, or image registration, lacking overall coordination considerations. To solve this problem, this article proposes a fast target detection algorithm for unmanned aerial vehicle power inspection based on YOLOv8 heterogeneous image fusion (YOLOv8-CVIFB), which can detect power devices fused with infrared and visible. The proposed algorithm mainly consists of three steps. Firstly, the CAO-C2F method is used to register the infrared and visible images collected by unmanned aerial vehicles. Secondly, the registered infrared and visible images are fused using the RFN-Nest method. Finally, the fused images were detected using YOLOv8s with Bi-Level Routing Attention added, generating images with detection results. The experimental results show that the algorithm proposed in this paper can reach 79.6% mAP and process 80 frames per second, at the same time the method can achieve higher accuracy and processing speed with a small computational improvement, which outperformed than some other algorithms.