基于级联式Snappy-CenterNet的锥套目标检测算法  

Drogue Target Detection Algorithm Based on Cascaded Snappy-CenterNet

作　　者：杨乐 张怡 王国屹 李华 Yang Le;Zhang Yi;Wang Guoyi;Li Hua(Hydraulic and Actuation System Department,AVIC Jincheng Nanjing Engineering Institute of Aircraft Systems,Nanjing 211106,China;Navigation Research Center,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China)

机构地区：[1]中国航空工业集团公司金城南京机电液压工程研究中心液压与作动系统部,南京211106 [2]南京航空航天大学导航研究中心,南京211106

出　　处：《兵工自动化》2023年第1期16-21,32,共7页Ordnance Industry Automation

摘　　要：针对空中加油因场景光照变化、环境遮挡等情况造成的锥套目标识别精度低、实时性差的问题,提出一种基于级联式Snappy-CenterNet深度网络的锥套目标检测算法。在CenterNet网络的基础上,以HourglassNet为主干网络,改进其bottleneck结构并引入中心池化的方法,对整体的网络结构进行优化,通过级联式的网络提升整体检测精度。实验结果表明:该算法可实现在多种复杂场景下对锥套目标的可靠检测,检测结果的精确率与召回率均可达99%,位置精度与区域精度分别可达99%与96%,更新率可达33.68 Hz,满足空中加油近距视觉导航阶段对于锥套识别的指标要求。Aiming at the problems of low accuracy and poor real-time performance of drogue target recognition in aerial refueling caused by scene illumination changes and environmental occlusion,a drogue target detection algorithm based on cascaded Snappy-CenterNet deep network is proposed.On the basis of CenterNet network,HourglassNet is used as the backbone network,the bottleneck structure is improved and the centr al pooling method is introduced to optimize the overall network structure,and the overall detection accuracy is improved through the cascaded network.The experimental results show that the proposed algorithm can reliably detect drogue targets in a variet y of complex scenes,and the precision and recall of the detection results can reach 99%,the position accuracy and region accuracy can reach 99%and 96%,respectively,and the update rate can reach 33.68 Hz,which meets the requirements of aerial refueling near vision navigation for drogue recognition.

关 键 词：空中加油 锥套识别 深度学习 目标检测 级联网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]