【作者】 李秀华；

【导师】 吴文福；

【作者基本信息】 吉林大学 ， 农业电气化与自动化， 2014， 博士

【摘要】 粮食安全和社会和谐、经济持续发展、政治稳定息息相关。我国储备粮粮库地域面积广、规模大、仓房容量大，粮食数量的智能稽核一直是储备粮自动监管和稽核的关键，但由于技术的局限性，目前国内外尚未有效解决这一难题。本文针对目前粮库稽核方法耗费大量的人力和财力的问题，提出了基于激光测距技术的粮食数量无损检测方法，重点研究了粮食边界识别、粮食表面上激光光斑所在位置的三维坐标计算方法和整体粮堆的体积计算方法，设计和开发了一套多线程数据采集及处理软件系统。主要工作如下：(1)提出了通过BP算法识别粮食种类来间接识别粮食边界的方法本系统是利用粮食表面上点的三维坐标进行体积计算，所以要识别出粮食边界。①设计了能够识别粮食种类的双隐层BP网。在对BP网的结构及算法进行分析的基础上，结合本课题的需求，通过训练和测试确定了12-28-20-1的网络结构。针对本文的研究对象（粮堆）无法从谷物籽粒形状来识别粮食种类的问题，提出采用无背景仅有粮食的图像的颜色矩向量作为特征向量来训练和测试BP网，使之能够识别参与训练的粮食种类。实现了对玉米这种粮食的识别，实验表明正确识别率很高。②实现了玉米粮堆的边界识别。从有背景（粮仓仓壁）的玉米图像的一角开始，逐行依次提取3*3窗口的颜色矩特征向量，依次输入BP网识别各3*3窗口是不是玉米，如果是则3*3窗口各像素灰度值设置为0，否则设置为255。最终得到玉米区域是黑色的，背景为白色的粮食边界二值灰度图像。粮食边界提取的结果表明：本文提出的粮食边界识别方法可行。(2)研究了粮食表面上激光光斑所在位置的三维坐标计算方法对粮食边界内部的激光光斑位置的三维坐标进行计算，否则不计算。①为了确定光斑在粮食图像中的位置，采用平均背景模型法提取了含有光斑的粮食图像中的光斑。实验结果表明，利用平均背景模型法提取的光斑位置及形状与该光斑所在图像中的位置及形状一致，该方法正确。②为了确定光斑是否在粮食边界内部及Delaunay分解等工作，采用基于重心的曲线拟合亚像素中心定位算法提取了光斑中心的亚像素坐标。实验结果表明，此方法定位精度较高。③根据光斑中心亚像素坐标在已提取的粮食边界图像中相应位置的像素值判定光斑是否在粮食边界内部，像素值是0则在内部，保留该光斑并计算其位置的三维坐标；是255则在外部，舍弃该光斑。实验结果表明，光斑是否在粮食边界内部的判定结果准确。④对光斑所在位置的三维坐标提出了原创性的计算方法。对设备进行标定后，利用激光测距仪测得的距离和云台的旋转角度，以及空间点在世界坐标系各个轴上的投影关系，推导出了云台垂直旋转角度在两种情况下的光斑所在位置的三维坐标计算公式。实验结果表明，本文推导出的光斑位置的三维坐标计算公式正确。(3)实现了粮食体积计算将二维Delaunay分解创新性地应用到了粮堆的三维体积计算中。对融合在一幅图像中的一组粮食表面上的光斑进行二维Delaunay分解，分解后得到多个不重叠的三角形，这些三角形的顶点为光斑中心，与粮食表面上的光斑点一一对应，从而粮食表面被这些三角形分割开，每个三角形下面为一个五面体，利用每个三角形顶点的三维坐标计算其下五面体体积，累加所有五面体的体积即可得到粮食总体积。在实验室内，对规则容器内的玉米体积进行了计算（玉米体积是已知的），实验结果表明，体积误差满足粮食监管需求。(4)设计和开发了一套多线程数据采集及处理系统以VS2008为开发平台，C++语言为开发工具，设计和开发了一套多线程数据采集及处理系统，系统软件包括：外设控制、数据采集、粮食边界识别、光斑提取、光斑中心的亚像素坐标提取、光斑所在位置的空间点的三维坐标计算、Delaunay分解、粮食体积计算、文件管理等内容。本系统从对各设备的控制到数据采集都是通过串口完成的，无需人工到达粮仓现场，实现了粮食实物数量的非接触无损检测，体积误差在3%左右，能够满足粮食监管部门的管理需求。系统设备少，实现成本较低，能够大幅降低监管和稽核成本，具有广阔的应用前景。更多还原

【Abstract】 Grain security is closely related to social harmony, political stability, sustained economicdevelopment. In our country, the reserved grain warehouses are distributed in different regionsand have large scale and their capacity is large. Intelligent audit of grain quantity is the key ofthe automatic reserved grain supervision all the time. But due to the limitations of technology,present researchers have failed to solve this problem effectively at home and abroad.In this paper, in view of the current grain audit method consuming a large amount ofmanpower and financial resources, the nondestructive measurement method of grain quantitybased on laser ranging technology is presented. The research designs and develops amultithreading data acquisition and processing software system and the research emphasisincludes grain boundary identification,3D coordinates computation method of the position of thelaser spot on the grain surface and grain volume calculation. The major work is as follows:(1) A indirect identification method of grain boundary by using the BP algorithm to identifythe categories of grain is proposedIn this system, grain volume is calculated by use of the3D coordinates of the points on thegrain surface, so the grain boundary needs to be identified.①Double hidden layers BP network is designed to recognize the categories of grain. Onthe base of analyzing the structure and algorithm of BP network, according to the requirementsof this project, the12-28-20-1network structure is determined by training and testing. To theresearch object (grain pile) of this thesis, the categories of grain cannot be identified by the shapeof the cereal kernel. In view of the problem, the color moment vectors of the grain image withoutbackground is used for the feature vectors to train and test BP network so that BP network canidentify the categories of grain participating in training. The recognition of corn was realized, theexperiment shows that the correct recognition rate is high.②The boundary identification of corn pile is achieved. Extract line-by-line the colormoment feature vector of the3*3window beginning with the one corner of the corn image withthe background (granary wall) and input it into BP network in turns. BP network judge whether itis corn. If it is corn, the pixels’ gray values of the3*3window are set into0, otherwise255. In theend, binary gray grain boundary image is gotten in which corn region is black and the background is white. The result of grain boundary extraction shows that the grain boundary recognitionmethod proposed in this thesis is feasible.(2) The calculation method of the3D coordinates of laser spot location on the grain surfaceis studiedThe3D coordinates of the laser spot location in the grain boundary are calculated, otherwisenot.①In order to determine the spot location in the grain image, the spot in the grain image isextracted by mean background model method. The experiment result shows that the location andshape of the spot extracted by mean background model method in the spot image are the same asthe location and shape of it in the grain image, so the method is correct.②In order to determine whether the spot is in grain boundary and Delaunay triangulation,the sub-pixel coordinates of the spot center are extracted by curve fitting sub-pixel centerlocalization algorithm based on gravity. The experiment results show that the positioningprecision of this method is higher.③Whether the spot is in grain boundary is judged according to the pixel value of thelocation of the spot center sub-pixel coordinates in the grain boundary image extracted. If thepixel value is0, the spot is in grain boundary and kept to compute the3D coordinates of itslocation; if the pixel value is255, the spot is outside grain boundary and is abandoned. Theexperiment result shows that the judgment that the spot is or not in grain boundary is accurate.④The original calculation method of3D coordinates of the spot location is presented.After the devices are calibrated, by use of the distance measured by the laser rangefinder, therotation angle of the PTZ and the projection of the space point on each axis in the worldcoordinate system, the formulas were derived to calculate the3D coordinates of the spot locationwhen the head’s vertical rotation angle is under four circumstances. The experiment results showthat the calculation formulas of the3D coordinates of the spot position are correct.(3) The grain volume calculation is realizedThe two-dimension Delaunay triangulation is innovatively applied to the3D volumecalculation of grain pile. After a group of spots merged in an image and on the grain surface aredecomposed by two-dimensional Delaunay triangulation, multiple triangles that don’t overlapare gotten. The triangle vertices are spot centers that they correspond to the spots on the grain surface. So the surface of grain is segmented by the triangles, and it is a pentahedron under everytriangle. The volume of the pentahedron is calculated by use of the3D coordinates of the verticesof each triangle, and the sum of the pentahedrons is the total volume of grain. In the laboratory,the volume of corn in the regular container is calculated (the volume of corn is known), theexperiment result shows that the volume error meets the demand of grain supervision.(4) A multithreading data acquisition and processing system is designed and developedTaking VS2008as the development platform, C++language as the development tool, amultithreading data acquisition and processing system is designed and developed. The softwaresystem includes: device control, data acquisition, grain boundary identification, spot extraction,spot center sub-pixel coordinate extraction, the calculation of3D coordinates of the space pointcorresponding to the spot location, Delaunay triangulation, grain volume calculation, filemanagementand so on.In this system, device control and data acquisition is achieved through the serial port, andworkers aren’t needed to go into the granary, so the non-contact nondestructive measurement ofgrain quantity is realized. The volume error is about3%and can meet the managementrequirement of the grain supervision department. This system demands less equipment and itscost is lower, so the cost of supervision and audit can be reduced greatly. This system has broadapplication prospects.更多还原