空间机器人深度学习识别捕获部位的应用探讨  

Application of Capture Position Deep Learning Recognition for Space Rbobots

作　　者：李宏坤 时中[1] 胡天健[1] LI Hongkun;SHI Zhong;HU Tianjian(Beijing Institute of Tracking and Telecommunications Technology,Beijing 100094)

机构地区：[1]北京跟踪与通信技术研究所,北京100094

出　　处：《飞行器测控学报》2017年第6期399-407,共9页Journal of Spacecraft TT&C Technology

摘　　要：针对空间机器人对捕获部位识别方法的普适性、实时性和准确性等要求,提出了采用深度学习方法对空间机器人捕获目标的特征部位进行识别。通过比较分析方法、数据驱动方法等传统识别方法和深度学习方法的优缺点,发现深度学习方法对于解决空间机器人捕获部位识别问题具有显著优势。进一步分析了应用深度学习方法解决捕获部位识别问题的几个关键技术问题,为后续空间机器人在轨捕获目标的研究与实践提供了新的思路。Universality,accuracy and realtime performance are critical problems in recognizing capture position for a space robot.In this paper,we propose the application of a deep learning method to solve the above problems.After a review of the pros and cons between traditional methods such as analysis and data-driven method and deep learning method for the robot capture problem,we conclude that deep learning method has a great advantage.We also analyzed some key technique problems to obtain a good performance for the application of deep learning in capture position recognition and the results provide a new view to both of research and engineering of space robots for on-orbital capture.

关 键 词：空间机器人 在轨捕获目标 特征部位识别 深度学习 

分 类 号：V476.5[航空宇航科学与技术—飞行器设计] TP242.2[自动化与计算机技术—检测技术与自动化装置]