Real-Time Object Detection of Visible-Infrared Feature Fusion for UAV Platform

Object detection based on visible images has been widely used in the fields of autopilot, wilderness search and rescue, and civil security, etc. However, the object detection algorithms for UAVs in complex backgrounds suffers from the problem of low object detection accuracy. Visible light images retain the color texture information of the target, but the imaging is poor under weak light conditions; infrared images achieve imaging by detecting the thermal radiation of the object, which is less affected by weather conditions. In this study, a multi-sensor feature fusion object detection algorithm is proposed, which is mainly based on YOLOv11 architecture and adopts the feature fusion strategy of “double trunk +MEFA” to fuse the features of visible and infrared modes. The experimental results show that mAP@.5 and mAP@.5:.95 reach 0.800 and 0.537 on the RTDOD dataset, and the tested FPS value reaches 184.166.To sum up, the proposed method can still maintain the object detection performance while reducing the training parameters.