An anchor-free drogue tracking network based on Shufflenet V2
Jin Li, Yafeng Wu, Yue Wang, Yang Hu · 2025
To solve the issue of tracking the refueling drogue in aerial refueling, a lightweight anchor-free drogue tracking network is developed in this paper. It is based on Shufflenet V2 and belongs to the twin network framework. According to the translation invariance needs of the twin network in the feature extraction stage, along with the Shffulenet lightweight network, a feature extraction network for the refueling drogue is created. The overall step size of this feature extraction network is 8, and the receptive field of the output feature makes up $68.5 \%$ of the template image, which is in line with the design standard of the typical twin network. Following the idea of Ocean/Ocean++, an anchor-free architecture is adopted by the network. Dilated convolutions in different directions are employed to mimic the anchor frames with diverse aspect ratios preset in the anchor frame network, guaranteeing the network’s responsiveness to different scales. Tests reveal that the network strikes a good equilibrium between precision and real-time operation, fulfilling the design prerequisites of the drogue tracking network.