A Cascade Adaboost and CNN Algorithm for Drogue Detection in UAV Autonomous Aerial Refueling

Xiaobin Xu,Haibin Duan,Yanjie Guo,Yimin Deng
DOI: https://doi.org/10.1016/j.neucom.2019.10.115
IF: 6
2020-01-01
Neurocomputing
Abstract:To promote the combat capability of unmanned aerial vehicles (UAVs) in the future battlefield, the autonomous aerial refueling (AAR) technology becomes a challenging research issue. An accurate position relationship between the tanker and the receiver is significant for AAR. A novel drogue detection method is presented in this paper. The Adaptive boosting (Adaboost) and the convolutional neural networks (CNN) classifier with the improved focal loss (IFL) function are utilized to detect the drogue in complex environments. The sample imbalance during the training stage of the CNN classifier is solved by the IFL function. The PyTorch deep learning framework is employed to implement the software system with the graphics processing units (GPUs). Real scenario images with a mimetic drogue on the tanker are captured for training and testing dataset by the airborne camera on the receiver. The experimental results indicate that the presented algorithm can accelerate the detection speed and improve the detection accuracy.
What problem does this paper attempt to address?