基于机器视觉的千粒质量测量仪的设计与试验  

Design and test of 1000-grain weight rapid measuring instrument based on machine vision

作　　者：胡正方 向阳[1] 熊瑛[1] 林洁雯 李亚军[1] HU Zhengfang;XIANG Yang;XIONG Ying;LIN Jiewen;LI Yajun(College of Mechanical and Electrical Engineering,Hunan Agricultural University,Changsha,Hunan 410128,China)

机构地区：[1]湖南农业大学机电工程学院,湖南长沙410128

出　　处：《湖南农业大学学报（自然科学版）》2021年第4期476-481,共6页Journal of Hunan Agricultural University(Natural Sciences)

基　　金：湖南省重点研发项目(2019NK2151);湖南省教育厅重点项目(19A224)。

摘　　要：针对目前作物千粒质量测定中数粒耗时长、精度低、自动化程度低、不适应多品种计数等问题,基于机器视觉技术,设计了一种集称重和数粒为一体的千粒质量测量仪。测量仪主要由机架、摄像头、载物盘、称重传感器、控制器等部件组成。工作时,通过称重传感器获取种子质量,同时通过摄像头采集种子图像并上传到上位机进行图像分析处理,获取种子数量,最后通过自动计算获取种子千粒质量并将结果显示到软件界面,以便及时获取测量结果。为了实现种子图像中粘连种子的分割,获取精准的种子数量,首先通过去边框处理提取种子区域,再采用中值滤波、最大类间方差法和形态学运算对种子区域进行预处理,采用基于距离变换的分水岭算法分割图像,通过统计连通区域的数量获得种子数量。对油菜、玉米、水稻和黄豆种子分别进行100组数粒、称重和千粒质量的测定,结果表明,作物种子数粒误差均不大于1粒;称重准确率分别为99.86%、99.92%、99.86%、99.96%;千粒质量测定的准确率分别为99.93%、99.63%、99.61%、99.97%;测量1次作物千粒质量仅耗时约2 s。In order to solve the problems of long time-consuming,low precision,low degree of automation,and inadaptability to multiple variety counting in the current crop 1000-grain weight measurement,a 1000-grain weight measuring instrument with the function of weighing and counting was designed based on machine vision.The device is mainly composed of a frame,a camera,a loading plate,a weighing sensor,a controller and other components.At the same time,seed images are collected by the camera and uploaded to the upper computer for image analysis and processing to obtain the number of seeds.Finally,the mass of thousands of seeds is obtained by automatic calculation and the results are displayed on the software interface so that the operator can obtain the measurement results in time.In order to realize the image segmentation of adhesive seed to get accurate seed number,the seed images were processed and analysis with extracting the seed region by removing the border,preprocessing the seed region by using median filtering,maximum inter-class variance method and morphological operation,and finally segmenting the image by using the watershed algorithm based on distance transformation,and obtaining the seed number by counting the number of connected regions.With rapeseed,maize,rice and soybean as the experimental subjects,100 groups of counting,weighing and 1000-grain weight determination experiments were carried out.The results showed that the number error was no more than 1 grain.Weighing accuracy was 99.86%,99.92%,99.86%,99.96%,respectively.The accuracy of 1000-grain quality were 99.93%,99.63%,99.61%,99.97%,respectively.The time of 1000 grain mass determination was less than 2 s.

关 键 词：机器视觉 千粒质量测量仪 分水岭算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]