基于视觉-运动双向伺服的田间品种选育机器人眼手协同仿生设计

Crop breeding is the focus of global agricultural hi-tech companies since to some extent it determines the national food safety. Automated crop variety screening technology faces up to the harsh inspection environment such as crop closing, leaf occlusion, and overlaps..In order to deal with the in-situ measurement demands during the whole crop growth period, a bio-inspired design methodology of the robotic eye-hand coordination operation is proposed for the crop breeding field robot based on the vision-motion mutual servo mechanism, which aims to make the robot autonomously remove the obstacles and masks, acquire images from an optimum view angle, and quantitatively compare the results among multiple measurements. Principally this research includes (1) completing mechanical design of the robot end effectors while mimicking manual operations for removing the overlap and occlusion and discovering the constitutive relations between the end effectors and the plants;(2) building models for the motion-control strategy and machine vision view angle configuration, constructing the vision-motion mutual servo algorithm while imitating the experts’ behavior and investigating the underlying biological mechanism;(3)realizing the object registration and quantitative representation for multi-angle multiple measurement for plant traits with the data acquired by the crop breeding robot while mimicking human recognition capacity, modeling the complex geometrical features of the plant traits and registration indices, and developing in an optimum view angle criteria and multi-image registration algorithm in an image feature space. A crop breeding field robot is to be built as per the above-mentioned techniques and methodology, and related theoretical and application achievements lay a foundation stone for measurement and control theories of the field agricultural robots.

种业事关国家粮食安全，品种选育是全球种业竞争焦点。农作物品种自动化优选需应对田间植株封垄、叶穗交叠遮蔽难检测场景挑战。.针对作物品种选育田间原位测量需求，提出基于运动-视觉双向伺服的品种选育机器人眼手协同作业仿生设计方法，实现自主去除遮挡、最佳角度成像、多次测量对比遴选。主要研究（1）模仿人工拨禾见穗去遮挡动作，揭示机器人末端-植株互作本构关系，实现末端仿生机械创成；（2）模仿专家寻找最佳检测视角行为，阐明生物学原型运动-视觉双向伺服机理，建立眼手系统移动策略与检测位姿控制模型，构建视觉-运动双伺服控制算法框架；（3）模拟人工多视角检测认知方法，构建植株表型复杂几何形貌表征和对标模型，在图像特征空间建立眼手系统最优视角判据和多次成像配准算法，实现“眼-手协同”机器人多角度多次检测对标识别及精细度量。基于上述研究研制田间品种选育机器人，相关成果将为我国农业机器人田间测控奠定装备技术基础。

作物生物特征识别是面向育种的高通量表型分析关键技术，动态自然场景加剧了“根-穗-茎-叶-粒”的参数自动化测量的挑战。本项目针对全生长周期内株叶穗型原位测量需求，提出基于运动-视觉双向伺服的品种选育机器人眼手协同作业仿生设计方法，引入了人机交互的有效方法，实现人机交互自然示教方法去除遮挡、最佳角度成像、多次测量对比遴选。主要研究了（1）基于机器人末端-植株互作本构关系，基于结构仿生原理实现末端仿生机械创成，研制了新型水稻分蘖激光投射检出器设备、类人型田间表型检测主动抗遮挡系统，建立了非全窥条件下的目标重建方法，实现了人工拨禾见穗去遮挡动作的模拟和部分可见目标物的高精度重建；（2）研制了顶视视角补全机器人，模仿专家寻找最佳检测视角行为，阐明生物学原型运动-视觉双向伺服机理，根据眼手系统移动策略与检测位姿控制规律建立了视角控制策略，构建视觉-运动双伺服控制算法框架；（3）模拟人工多视角检测认知方法，构建植株表型复杂几何形貌表征和对标模型，在图像特征空间建立眼手系统最优视角判据和多次成像配准算法，实现“眼-手协同”机器人多角度多次检测对标识别及精细度量。实验研究表明，对精细小尺度生物特征，提出的表型检测策略具有较高的有效性。基于上述研究研制了2款田间品种选育机器人(类人型和环绕观测型)，重点突破了基于作物生理特征主动智能识别的高通量表型量化技术，相关成果为农业机器人田间测控及表型装备系列化奠定装备技术基础。