In order to realize plant morphological measurement and physiological diagnosis, a multi-modal three-dimensional (3D) reconstruction method was proposed. This reconstruction method further laid the foundation for plant phenotypic measurement. Due to the complexity of 3D geometric morphologies, only two or two-and-a-half-dimensional images of greenhouse plants can be captured at a single angle of view (AOV) by the imaging sensors. However, 3D point cloud reconstruction of plants requires images captured at multiple AOVs. In addition, the 3D geometric morphologies of plants undergo significant changes during the full-growth-cycle and to acquire suitable 3D plant images, it is necessary to frequently adjust the sensor position. Therefore, sensor position and AOV directly affect the plant phenotyping efficiency. Developing an efficient and accurate multi-view 3D point cloud reconstruction method that meets the need for full-growth-cycle, high-throughput 3D reconstruction and phenotyping of greenhouse plants is therefore pivotal to the development of high-throughput plant phenotyping techniques. So a multi-modal three-dimensional reconstruction method of greenhouse tomato plants under different measurement positions and angles was proposed, and to solve the problem of multi-spectral reflectance mapping and multi-view point cloud 3D reconstruction, multi-spectral reflectance images were registered to RGB-D image coordinate system by phase-correlation method, and a multi-view RGB-D image 3D reconstruction method based on self-calibration of the Kinect sensor was established which realized the reconstruction of RGB 3D point cloud model and multi-spectral reflectance 3D point cloud model of the plants. The two-dimensional multi-spectral image registration quality was evaluated objectively by the normalizing gray-scale similarity coefficient D, the spectral overlap rate in the region of interest (ROI) C, and the mutual information value and the Hausdorff distance HD was applied to objectively evaluate the reconstruction accuracy of the three-dimensional point cloud reconstruction of the plant. In total, 30 greenhouse tomato plants were used in this study with each plant reconstructed from four angles of view at angle intervals of 90 degrees. The obtained results showed that the average values of C and D were 0.920 6 and 0.908 5, respectively. After registration, the mutual information value increased by 9.81 % and the canopy multi-spectral images could be registered accurately to the depth coordinate system. The ratio of the HD distance set of reconstruction point cloud less than 0.6 cm was 78.39 %, the ratio of less than 1.0 cm was 91.13%, and the mean value of the tomato distance ensemble HDavg was 0.37 cm, depicting that the tomato plant 3D point cloud model had high reconstruction accuracy and could be applied to multi-modal 3D reconstruction of greenhouse tomato plants. This research integrated the traditional 3D geometric morphology measurement system and plant physiological information diagnosis system, and as such the external morphology and internal physiological information of the plants could be measured in the same imaging room. It provides a precise and efficient measurement method for high-throughput plant phenotypic measurement, and is of great significance to the development of modern intelligent facility horticultural management and plant phenomics.

为实现冬小麦田间精准变量施肥,针对自行研制的基于光谱技术的冬小麦追肥机械的控制特性,构建双变量控制模型,研究确定系统最优控制策略.采用Bisquare估计稳健回归分析变量施肥机构的控制模型,在此基础上,从转速优先控制、开度优先控制以及双变量自适应控制3个方面研究分析变量施肥控制策略.为了兼顾变量施肥执行效率与控制精度,最终构建变量施肥控制序列查询表,通过查表法来实现双变量自适应控制.结果表明,通过查表法可以快速地获得目标转速和开度的控制量,然后将其输入模糊PID控制器中,运行模糊PID控制算法能够实现双变量精确控制,其中转速控制误差低于12.83%,平均误差小于9.84%;开度控制误差低于13.57%,平均误差小于9.34%.由此可见,本研究构建的变量控制模型与策略性能良好,可以满足冬小麦精准变量施肥的技术要求.

为实时获取浅根系作物的根系生长形态,设计了一种可用于多点测量的微型根系形态实时原位采集系统.系统主要由微型摄像头和光学放大元件等组成(体积1.5 cm3),采集的图像通过无线模块发送至终端.采用基于区域生长的根系图像分析方法,以腐蚀图像为出发点,膨胀图像为终止点,结合相似性准则进行区域生长、区域标记和区域保留,来滤除土壤孔隙和杂质等对图像产生的干扰,从而提取根系轮廓,并通过图像形态学计算得到根长密度、根系平均直径等形态参数.以此系统采集樱桃番茄、辣椒根系形态参数,试验结果表明,根系长度测定值的绝对误差不超过1.5 mm,相对误差不超过5.3％;根系平均直径绝对误差不超过0.09 mm,相对误差不超过6.7％.与土壤采样法测定值相比,在0～10、>10～20、>20～30和>30～40 cm 4个土壤层内2种测定方法根系平均直径决定系数R2>0.87(P<0.01),根长密度在30 cm深度以内的土壤层决定系数R2>0.81(P<0.01).证明本文设计的微型根系形态实时原位采集系统具有较高的准确性,可用于浅根系作物形态的多点观测.

[目的]设计一种能在作物行间自主导航的施药机器人,实现移动机器人在温室中自动行走并均匀施药.[方法]针对导航路径识别受光线变化影响较大的问题,在Kinect摄像机获取的彩色图像中选取了HIS空间,并对K-means算法的聚类中心和聚类数目的选取进行了优化,随后采用改进的K-means算法对与光照信息无关的H、S分量联合分割,获得完整道路信息,并采用Candy算子检测边缘及改进的Hough变化方法拟合导航路径.采用模糊控制方法通过实时调整转角和转向,对车体行走偏移进行矫正.同时,为满足不同农作物的施药需求,在喷药系统上选用了自整定模糊PID控制算法.[结果]该系统可有效适应不同光照条件,提取作物行中心线平均耗时12.36 ms,导航偏差不超过5 cm,植株叶片正面的上、中、下层覆盖率分别为63.26％、50.89％和75.82％,单位面积(1 cm2)雾滴数平均为55、42和78个.[结论]本系统可以满足温室移动机器人自主施药防治病虫害的需求.

Based on Hilbert transformation, a pest counting system and remote publishing system were designed and established. The system was composed of pest–killing lamp, acquisition card, voltage transformer, PLC, and router, etc. Under operation, the voltage from pest–killing lamp was reduced by voltage transformer, and then collected by acquisition card, which connected with PC by USB. LabVIEW was used to complete the Hilbert process to obtained the instantaneous value. When the instantaneous was less than the set value, modbus instruction was sent to PLC serial port, and PLC will count at receiving corresponding instruction. PLC and MySCADA were introduced to attain remote publishing on mobile phone and computer. Disturbance information of 714 pests were analyzed, and totally 4 terms of 7 disturbance information were listed in the paper. The result showed that the system could precisely identify 7 disturbance area, and display by mobile phone and remote computer.

基于Kinect体感感应技术,设计了一套温室果蔬采摘运输自动跟随平台.通过体感感应系统获取图片上像素点的深度信息,结合图像处理算法,逐行扫描确定人体图像并实时获取人体骨骼信息,计算了人体当前的三维坐标并记录人体走过的路径轨迹.系统采用自调整函数对路径进行优化,避免了剧烈转向行为,并对优化后的路径以模糊算法动态确定纯追踪模型的前视距离,从而实时调整转向和转角,实现了精准跟随和稳定跟随.试验结果表明,该跟随系统能在避免剧烈转向的前提下以较高的精度跟随,横向最大跟踪偏差不超过10.0 cm,最大深度偏差为5.5 cm,且系统性能不受光照条件影响,满足温室采摘运输要求.

针对春冬季节温室内CO2浓度低下的问题，采用风送式CO2气体补偿装置，调控温室环境CO2浓度，提高光合作用速率。以补偿时间为输入量，补偿效果和补偿速率为输出量，建立补偿时间和补偿量的线性关系，通过定时定压的“多次少量施放，超过上限停施”的方式向温室补充CO2，调控温室环境中的CO2含量，以达到精确补偿的效果。结果表明，向温室一次补偿5 min的CO2气体，室内CO2平均浓度在15 min内由205μmol/mol达到540μmol/mol，间歇补偿3次，室内 CO2平均浓度在45 min内由205μmol/mol 达到1200μmol/mol。说明补偿时间与其补偿量呈线性关系，标准偏差在0~3.03%范围内，实现了温室CO2快速精确补偿的功能。

为了实现冬小麦实时变量精确追肥,研究了基于模糊PID控制技术的变量追肥机追肥量实时调整算法,通过对排肥器转速和开度双变量调节,实现追肥量的优化控制.系统首先采用粒子群优化算法确定PID控制器参数的初始值,然后通过实时获取冬小麦冠层归一化植被指数和排肥器实时状态,结合模糊控制理论和PID控制技术,对PID参数进行在线整定,实时调整排肥器的转速和开度,从而实现追肥量的最优控制.试验结果表明:施肥过程中,施肥量存在波动性.但施肥量变异系数小,最大为3.22％,均值为2.09％,可以满足田间变量施肥的要求.模糊PID控制算法具有良好的动态稳定性和跟踪性能,无论是室内试验还是大田试验的控制精度均达到86％以上.

In view of the shortcomings of cable-driven and chain transmission mechanism with motor rotation direction changing frequently and large load,a novel electric-drive robot leg based on planar fivebar mechanism with dual freedoms was proposed to reduce the inertia of robot leg and enhance load capacity. Considering the close relationship between the dynamic performance of robot and the dimension parameters of planar five-bar mechanism,the kinematic and dynamic analyses of this mechanism was made,and the functional relationship between peak torque,peak angular velocity of joint motors,energy consumption in a gait period and dimension parameters were setup to get the optimal dimension parameters. In this way,a multi-objective optimization model of the dimension parameters was got. By determining the weight of the objectives based on analytic hierarchy process( AHP) method,the multiobjective optimization problem was transformed into single objective optimization problem. Then the genetic algorithm was applied to obtain the optimal solution. Finally,the virtual prototypes of robot leg before and after optimization were built via ADAMS to conduct the walking simulation test,and the comparison between simulation results and computation results was made. The motors were selected subsequently according to the optimization results,which made the robot weight decreased by 7. 8%,which was beneficial to the improvement of load capacity and battery life. These results verified the correctness and validity of the proposed algorithm which can also be applicable to the selection of brushless DC motor.

[目的]考查在温室中采用不同水压下喷雾降温的效果.[方法]以可编程逻辑控制器、变频器、水泵、触控系统和多路传感器为硬件平台,设计了温室变频雾化装置.试验探索5个不同喷雾频率下,系统的实际降温效果,并在此基础上提出了适合试验温室的喷雾压力.[结果]喷雾压力低于110 kPa(对应喷雾频率为31 Hz)时,降温幅度随着压力的增加而升高,但当喷雾压力大于110 kPa时,随着压力的增加降温效果趋于平滑,而温室内湿度随着压力的增加不断增加,喷头的流量随着压力的升高而近似线性增加.[结论]综合考虑温度和湿度2个因素,该试验温室的最适应喷雾为压力为100～120 kPa.