【作者】 李凯；

【导师】 崔永杰；

【作者基本信息】 西北农林科技大学 ， 农业机械化工程， 2024， 博士

【摘要】 猕猴桃作为高度依赖授粉的果品经济作物,能否对其花朵进行充分有效授粉,极大程度决定了猕猴桃的座果品质以及猕猴桃的产量。然而,就目前的自动化授粉研究而言,需要解决如何提高花粉利用率等问题。因此,本文分别从猕猴桃花朵精准授粉关键参数优选研究入手,探寻了授粉末端喷授参数的获取方法,进行了猕猴桃授粉装置总体方案、授粉机械臂的仿真设计,开展了猕猴桃花朵定位及花簇位姿判别,精准授粉控制方法以及猕猴桃授粉装置试制等方面的研究。所开展的有关猕猴桃的精准对靶授粉方法研究,旨在根据花朵位姿确定猕猴桃授粉装置末端喷头的作业位姿,凭借对靶授粉,实现对花粉的高效利用,减少花粉使用量,最终通过研制符合猕猴桃种植农艺特点的授粉机械臂,在简化机械结构和作业动作的同时,确保能够精准对靶授粉。主要研究内容和结论如下:（1）猕猴桃花朵精准授粉关键参数优选。根据猕猴桃花朵自动授粉作业过程中存在的四种情况,开展了授粉末端执行器的喷雾特性研究,分析了不同情况下的授粉模式特点,搭建了喷雾特性研究试验平台,利用其验证了不同授粉模式下,授粉距离和环境风速对授粉效果的影响。不仅证明了16cm授粉距离和0.24L/min授粉流量等最佳授粉参数的有效性,而且根据实际风速和花朵位姿偏角来调整喷头位姿角度方案的可行性,即便花朵存在位姿偏角或有风影响,但只要风速不超过2.5m/s,亦可通过修正喷头的位姿偏角来减缓风力及位姿偏角对喷雾效果的影响。（2）猕猴桃授粉装置总体设计方案。为了提高自动授粉装置的授粉效率和授粉精度,研究并设计试制了用于自动化连贯授粉的授粉机械臂,基于棚架栽培特点,仿真设计了符合授粉作业要求的6自由度授粉机械臂。先根据棚架空间大小,确定了机械臂的6自由度构型、尺寸参数以及作业空间,然后,通过仿真,分析了用于构建肩部、肘部支撑件与转动件的尼龙材料的应力应变强度,最后,通过MATLAB,仿真验证了机械臂在以最大速度ω_max=0.105rad/s移动作业时,末端喷头能以任意位姿在750mm×750mm×200mm的作业空间范围内移动,且保证各关节的速度和加速度不发生突变。由此验证了授粉机械臂作业空间,关节结构的设计合理性,以及轨迹规划和作业运行的稳定性。（3）猕猴桃花朵定位及花簇位姿判别方法研究。为了能够有效判别猕猴桃花朵位姿,先采用SOLOv2模型定位猕猴桃花簇图像中的猕猴桃单花目标,并在此基础上,采用Mobile Net V2模型实现了猕猴桃单花的关键点检测,在此基础上由关键点绘制出猕猴桃单花的位姿预测线;通过计算图像中猕猴桃花朵之间的交并比（Io U）,以确定单花所形成的花簇,同时取花簇中各猕猴桃单花位姿预测线的平均值,作为整个猕猴桃花簇的位姿预测线,从而实现了对复杂背景环境当中的花簇位姿判别。结果表明:SOLOv2模型处理猕猴图花簇图像时,其平均精度AP、Io U阈值为0.50的掩码平均精度AP₅₀和Io U阈值为0.75的掩码平均精度AP₇₅分别为61.70%、84.90%和63.30%,分割速度为0.183s/幅,性能明显高于Mask Rcnn、SOLO和YOLACT三种模型。花朵位姿被有效预测的概率为74.20%。（4）猕猴桃授粉装置精准授粉控制方法研究。基于花朵位姿信息,研究耦合花朵位姿的机械臂末端位姿调控策略,提出利用花朵位姿来确定末端喷头授粉标准姿态的精准对靶授粉思路及策略。同时,在ROS（Robot Operating System,机器人操作系统）中使用Move It!（注:Move It由ROS中一系列移动操作的功能包组成,包含运动规划,操作控制,3D感知,运动学,碰撞检测等等,是目前针对移动操作最先进的软件。）对机械臂进行路径规划,集成实体机械臂,通过仿真分析和试验验证了路径规划的可行性。采用三维建模软件Solidworks创建了6自由度授粉机械臂模型,通过sw2urdf插件创建了有关授粉机械臂模型的URDF文件。基于ROS对6自由度实体机械臂进行运动控制,选取开源运动规划库（OMPL）中的随机扩展树算法（RRT）进行授粉机械臂的路径规划。并利用MATLAB进行机械臂运动学仿真,验证机械臂工作空间满足预期要求。基于ROS,建立机械臂与相机的通信,集成搭建猕猴桃授粉装置,并开展有关控制系统的准确性测试,其授粉花朵的实际案例表明:为授粉坐标为（0.123508,-0.124471,0.581000）的猕猴桃花朵,在控制系统对机械臂进行规划的过程中,1轴电机转动32.4°;2轴电机转动-18.2°;3轴电机转动-21.8°;4轴电机转动0°;5轴电机转动-3.8°;6轴电机转动0°,基本能够控制机械臂末端移动至目标位置处。（5）猕猴桃授粉装置试制及验证试验。根据上述的研究成果,集成研制了适用于棚架栽培环境的猕猴桃授粉装置,并开展了相关验证试验,同时通过猕猴桃授粉装置验证了猕猴桃花朵位姿判别以及精准对靶授粉方法。结果表明:猕猴桃授粉装置的花朵检测准确率为92.44%,有效授粉率为87.44%,授粉装置信息感知系统处理所拍照猕猴桃花朵图像的平均时间约为1s/幅,验证了本研究所提出的精准对靶授粉的有效性。其果园验证试验表明:猕猴桃授粉装置的花朵检测准确率为86.53%,有效授粉率为81.43%。更多还原

【Abstract】 Kiwi,a highly pollination-dependent fruit,crop yield and fruit quality largely depends on the efficient and effective pollination of its flowers.However,the research in robotic arm automation for pollination presently needs to solve problems such as improving the utilization rate of pollen and creating machinery.Therefore,this dissertation studied the optimization of key parameters for precise pollination of kiwi flowers,the method for obtaining optimal parameters for the pollination end spray,the overall scheme of the kiwi pollination device,the simulation design of pollination robotic arm,positioning of kiwi flowers,and clustering of the pose detection.It also studied the precision pollination control method and the trial production of the kiwi pollination device.The research aims to determine the working position of the end nozzle of the kiwi pollination device according to the pose of the flower,thereby achieving efficient utilization of the pollination by targeting;reducing the amount of pollen used,and developing a pollination robotic arm that meets the agronomic characteristics of kiwi cultivation.This research ensure precise target pollination while simplifying mechanical structure and operational actions.The main research content and conclusions are as follows:(1)Study on the optimization of key parameters for precise pollination of kiwi flowers.According to the four situations that exist in the automatic pollination operation process of kiwi flowers,research was carried out on the atomization characteristics of the pollination terminal actuator.The pollination mode features and pollination mechanisms under different situations were analyzed.The atomization characteristic test platform was set up,and it verified the effect of pollination distance and environmental wind speed on pollination effectiveness under different pollination modes.Based on the actual wind speed and flower pose deviation angle,the feasibility of adjusting the nozzle pose angle scheme was verified,proving that even if there is pose deviation in the flower or wind effects,the effect of wind force and pose deviation on atomization may be alleviated by correcting the pose deviation angle of the nozzle,as long as the wind speed does not exceed 2.5m/s.(2)Overall design scheme of kiwi pollination device.To improve the pollination efficiency and precision of the automatic pollination device,a pollination robotic arm was designed and developed for automated and coherent pollination.The agronomic characteristics of kiwi cultivation were taken into account,and simulation was performed to design a 6-degree-of-freedom robotic arm suitable for pollination.The robot arm’s 6-degree-of-freedom configuration,size parameters,and operating space were determined based on the size of the space.The stress-strain strength of the nylon material used to construct the shoulder and elbow supports and turning parts was analyzed through simulation,and the mechanical arm was simulated and verified using MATLAB.(3)Study on kiwi flower positioning and flower cluster pose detection method.An effective judgment method for the pose of kiwi flowers was developed using the SOLOv2 model to determine the pose prediction line of single flowers in the kiwi flower cluster images;the Mobile Net V2 model was then used to detect the key points of a single kiwi flower,and the pose prediction line of the single kiwi flower was drawn from these key points.The intersection over union(Io U)of the kiwi flowers in the image was calculated to identify the flower clusters formed by the single flowers,and the average of the pose prediction lines of each single kiwi flower in the flower cluster was taken as the pose prediction line of the entire kiwi flower cluster,realizing flower cluster pose detection in complex background environments.(4)Research on the precise control method for the kiwi pollination device.Based on the pose information of the flowers,a strategy for controlling the end position of the robotic arm linked to the flower pose was studied.A precise targeting pollination approach was proposed using the position of the flowers to determine the standard stance of the end effector for pollination.In the process,the Robot Operating System(ROS)employed Move It! for path planning of the robotic arm,integrating both virtual simulations and physical trials to validate the feasibility of the planned paths(Note: Move It consists of a series of functional packages for mobile operations in ROS,including motion planning,operation control,3D perception,kinematics,collision detection,etc.It is currently the most advanced software for mobile operations.).Utilizing the 3D modeling software Solidworks,a 6-degree-of-freedom robotic arm model was created,and the URDF(Unified Robot Description Format)file of the model was generated using the sw2 urdf plugin.Motion control of the physical 6-degree-of-freedom robotic arm was conducted within ROS,employing the Randomly-Exploring Random Tree(RRT)algorithm from the Open Motion Planning Library(OMPL)for the path planning of the pollination arm.MATLAB was later utilized for the kinematical simulation of the arm,confirming that the working space of the robotic arm met the anticipated requirements.Based on ROS,establish communication between the robotic arm and the camera,integrate and build a kiwifruit pollination device,and conduct accuracy testing on the relevant control system.The actual case of the pollinated flower shows that for kiwifruit flowers with pollination coordinates of(0.123508,-0.124471,0.581000),during the planning process of the robotic arm in the control system,the 1-axis motor rotates 32.4 °;2-axis motor rotation-18.2 °;3-axis motor rotation-21.8 °;4-axis motor rotates 0 °;5-axis motor rotates-3.8 °;The 6-axis motor rotates 0 ° and can basically control the end of the robotic arm to move to the target position.(5)Trial production and validation experiments of kiwi pollination device.An integrated kiwi pollination device suitable for greenhouse cultivation was developed based on the above research results.Relevant validation experiments were carried out,and the device verified the kiwi flower pose detection and precision targeting pollination methods.The results showed that the accuracy of flower detection in the kiwifruit pollination device was 92.44%,and the effective pollination rate was 87.44%.The average time for the information perception system of the pollination device to process the kiwifruit flower images taken was about 1 second per image,which verified the effectiveness of the precise target pollination proposed in this study.The field validation test showed that the accuracy of flower detection in the kiwifruit pollination device was 86.53%,and the effective pollination rate was 81.43%.更多还原