级联特征点检测的锥套小目标快速检测定位算法

为提升无人机自主空中加油中锥套小目标的检测精度和实时性，提出了级联网络与特征点检测网络的锥套小目标高精度快速定位算法。该算法设计了锥套目标全局粗定位和局部精定位的两级检测神经网络结构，采用特征点的输出位置误差及特征点群拟合的椭圆参数误差设计网络的损失函数，利用特征点拟合的锥套目标尺寸位置信息修正跟踪算法，提升了跟踪算法的和目标定位的实时性。实验测试结果表明，该定位算法定位成功率在95%以上，定位的精度（预测区域与真实区域的重叠率）在80%以上，定位输出速率达到100 Hz，对于环境的变化有着较强的适应性。因此该算法可以快速准确地进行锥套目标的跟踪，对于无人机空中加油技术的发展具有重要的研究意义。

Fast detection and location algorithm for small drogue based on cascaded feature point detection

In order to improve the detection accuracy and real-time performance of small drogue in autonomous aerial refueling of UAV, this paper proposed a fast and accurate location algorithm for small drogue based on cascade network and feature point detection network. It designed a two-stage detection neural network structure for global rough localization and local precise localization of drogue, and used the output position error of feature points and the elliptic parameter error of feature group fitting to design the loss function of the network. The method used the size and location of the drogue to correct tracking algorithm. The experimental results show that the proposed algorithm has a successful localization rate of more than 95%, a localization accuracy of more than 80%, and a localization output rate of 100 Hz. It has a strong adaptability to the changes of the environment. In summary, the algorithm can track the drogue target quickly and accurately, and has important research significance for the development of UAV aerial refueling technology.