基于深度神经网络的自适应图像伺服空中对接控制  

Adaptive image-based visual servo aerial docking control based on deep neural network

作　　者：黄远潭 任锦瑞 刘润潇 全权[1] HUANG Yuantan;REN Jinrui;LIU Runxiao;QUAN Quan(School of Automation Science and Electrical Engineering,BUAA,Beijing 100083,China;School of Automation,XUPT,Xi’an 710100,China)

机构地区：[1]北京航空航天大学自动化科学与电气工程学院,北京100083 [2]西安邮电大学自动化学院,陕西西安710100

出　　处：《飞行力学》2025年第1期63-71,共9页Flight Dynamics

基　　金：国家自然科学基金资助(62103335);陕西省科学技术协会青年人才托举计划项目资助(20230111)。

摘　　要：基于图像的视觉伺服是机器人视觉领域的研究热点之一,尤其是在无人机领域。为解决侧风等气流扰动对无人机图像伺服控制带来的影响,首先依据图像伺服基本模型,推导出小机动无人机图像误差变化率的扰动项;然后利用深度神经网络离线学习图像误差变化率的风扰动项,在此基础上利用在线自适应技术对速度控制器实时补偿。仿真结果表明,所提方法能使空中加油对接效率在侧风干扰条件下比常规的基于图像的视觉伺服提高约20%。Image-based visual servo is one of the research hotspots in the field of robotic vision,es-pecially in the realm of unmanned aerial vehicles(UAVs).To address the impact of aerodynamic disturbances,such as crosswind,on image-based servo control for UAVs,the first step involves de-riving the disturbance term for the rate of change of image error for a highly maneuverable UAV based on the fundamental model of image-based servo.Subsequently,a deep neural network is em-ployed to offline learn the wind disturbance term for the rate of change of image error.Building upon this,an online adaptive technique is utilized to compensate the speed controller in real-time.The simulation results indicate that the proposed method can enhance the efficiency of aerial refueling docking in the presence of crosswind disturbances by approximately 20%,compared to conventional image-based visual servo methods.

关 键 词：视觉伺服 图像伺服 空中对接 深度神经网络 自适应控制 

分 类 号：V279[航空宇航科学与技术—飞行器设计] TP183[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]