金保华 殷长魁 张卫正 张伟伟

中圖分类号：TN959.4;S758.4;TS255.1文献标识码：ADOI：10.3969/j.issn.2096-1553.2019.02.010

文章编号：2096-1553（2019）02-0071-11

关键词：机器视觉;苹果园;果实识别;图像处理;机器学习

Key words：machine vision; apple garden;fruit recognition;image processing;machine learning

摘要：从基于颜色阈值、形状和纹理的果实识别，三维果实形态识别，夜间果实识别，基于机器学习的果实识别，阴影和遮挡影响下的果实识别5个方面，对基于机器视觉的苹果园果实识别研究现状进行了综述，认为上述研究所涉及的算法较为复杂，功能也很强大.但鉴于视觉理论、图像处理技术和硬件条件等限制，以及苹果园复杂多变的环境，基于机器视觉的果实识别目前尚无理想的方法，未来的研究重点应包括：1）加强更有效的图像增强、图像分割和特征提取等算法的研究，有效解决果实重叠、遮挡、颜色和光线变化的影响;完善白天和夜间果园现场作业的识别算法，建成全天候作业采摘机器人.2）加强基于自监督学习的果实识别的研究，以增加模型接收的反馈信息和模型表征的复杂的适用任务类型，减少任务中涉及的人类手工劳动比重，提高自动化程度.3）加强图像的自动获取与果实识别的研究，结合计算机视觉与近红外、激光雷达等检测技术，集成多模态的图像和非图像信息进行果实识别，提高处理速度和实时性，以及识别的准确度及系统的稳健性，为苹果自动采摘、果园的精准管理提供借鉴.

Abstract：The current situation of fruit recognition based on machine vision was reviewed from fruit recognition based on color threshold， shape and texture， three-dimensional fruit shape recognition， nocturnal fruit recognition， fruit recognition based on machine learning， fruit recognition under the influence of shadow and occlusion. Its thought that the algorithms involved in the above research were more complicated and features were very powerful. However， in view of the limitations of visual theory， image processing technology and hardware conditions， as well as the complex and varied environment of apple garden， there was no more ideal technology for machine vision-based fruit recognition， and it needed to be improved. Future research focuses include： 1） Strengthening more effective algorithms for image enhancement， image segmentation， and feature extraction to effectively address the effects of fruit overlap， occlusion， color， and light changes; and improving the identification algorithms for day and night orchard field operations for the construction of an all-weather operation picking robot. 2） Strengthening the research on fruit recognition based on self-supervised learning to increase the feedback information received by the model and the complex applicable task types of model representation， reduce the proportion of human manual labor involved in the task， and improve the degree of automation. 3） Strengthening the research of automatic image acquisition and fruit recognition， combined with computer vision and near-infrared， laser radar and other detection technologies， integrating multi-modal image and non-image information for fruit recognition， improving processing speed and real-time， and identifing accuracy and system robustness to provide reference for apples automatic picking and precise management of orchard.

0 引言

苹果园果实识别是果实定位、自动化采摘和果树估产的基础.由于苹果树在自然状态下生长姿……