Deep Learning Approach to Drogue Detection for Fixed-Wing UAV Autonomous Aerial Refueling with Visual Camera

In the unmanned aerial vehicle (UAV) autonomous aerial refueling process, detection and position feedback of the refueling drogue by the refueling UAV are essential. Given the short docking distance and the characteristic of being easily affected by light, weather, and complex environments during the UAV refueling process, realizing real-time and robust drogue detection is a critical problem. In this study, we designed and implemented a set of tanker UAV-refueling drogue detection and recognition models based on the YOLOv5s model for real flight scenarios of fixed-wing UAVs. During the research process, the fixed-wing UAV test platform was built. Then, based on an actual flight test, 1900 flight images of the refueling drogue in various environments were collected, and a tanker UAV-refueling drogue dataset was produced. Subsequently, based on the YOLOv5s detection model, two detection models, YOLOv5s-D and YOLOv5s-DP, were designed and optimized for detecting tanker UAV and the refueling drogue. The performance of the fixed-wing UAV test platform was assessed using the detection model to complete the test flight verification in an actual scene. The experimental results showed that the mean average precision (mAP) of the drogue detection model carried by the refueling UAV was 97.2%, and the real-time detection frame rate of the tanker UAV and drogue was 33.5 fps, which validated the real-time performance of the detection model. Compared with previous research results, these findings provide advancements in image data acquisition of small fixed-wing UAV refueling drogues, the design of the target detection algorithm, and the flight test of drogue detection in air.