Abstract: Fruit picking robot is of great significance to realize automatic fruit picking, and vision system is the key to the research of fruit picking robot. Therefore, the research work on the key technologies of vision system of picking robot at home and abroad in recent years is summarized. According to the technical route of the picking vision system, the image acquisition technology of the picking vision system is discussed. This paper summarizes the common target recognition algorithms in the field of fruit target recognition, including singlestage algorithm, twostage algorithm, semantic segmentation and instance segmentation algorithms based on deep learning and their improvements, and summarizes the application of recognition technology in complex environment. The hardware system and software algorithm of fruit target location are summarized based on the method of picking point acquisition. Finally, the paper discusses the limitations of the picking vision system in algorithm, hardware and applicable environment, and proposes that the future research should focus on multiinformation fusion, deep learning technology and vision system in complex environment, so as to provide reference and guidance value for the research of fruit picking robots.

Key words: fruit, harvesting robots, machine vision, picking point localization, target recognition

摘要： 水果采摘机器人对实现水果自动化采摘具有重要意义，视觉系统是采摘机器人研究的关键。为此，对近几年国内外采摘机器人视觉系统的关键技术研究工作进行总结。按照采摘视觉系统的技术路线，论述采摘视觉系统图像采集技术；总结水果目标识别领域常用的目标识别算法：单阶段算法、二阶段算法、基于深度学习的语义分割和实例分割算法及其改进算法，总结在复杂环境下识别技术应用的关键；并以采摘点的获取方法为轴心，选择水果目标定位的硬件系统和软件算法。最后探讨采摘视觉系统目前面临算法、硬件、适用环境等方面的局限性，提出未来应该关注多信息融合、深度学习技术、复杂环境下视觉系统等方向的研究，为水果采摘机器人的研究提供借鉴和指导价值。

关键词: 水果, 采摘机器人, 机器视觉, 采摘点定位, 目标识别