基于单幅图像的玉米果穗数量性状测量方法  

Measurement method for quantitative characteristics of maize based on single image

作　　者：高启 李毅念[1] 金典 龚成杰[1] 范亮亮 李刘阳洋 GAO Qi;LI Yinian;JIN Dian;GONG Chengjie;FAN Liangliang;LI Liuyangyang(College of Engineering,Nanjing Agricultural University,Nanjing 210031,China)

机构地区：[1]南京农业大学工学院,江苏南京210031

出　　处：《南京农业大学学报》2018年第3期562-569,共8页Journal of Nanjing Agricultural University

基　　金：国家大学生创新创业训练计划项目(201610307096)

摘　　要：[目的]玉米果穗数量性状是玉米考种和产量预测的重要参数,主要包括穗行数、行粒数和总粒数。快速精确地测量果穗数量性状是玉米自动化考种的难点问题。[方法]选用干玉米果穗进行试验,以单幅果穗图像为研究对象,在改进YCrCb色彩空间的基础上利用双峰法提取果穗轮廓,将图像沿穗轴等分为20部分,对每部分采用像素累计法得到穗行的定位点,对定位点线性插值得到穗行线,通过分析穗行线上的像素变化得到行粒数,利用果穗横截面的几何模型估算穗行数,并采用补偿法校正单幅果穗图像的成像误差以提高其计算精度,最后将行粒数与行数相乘得到总粒数。[结果]该方法对光照环境和图像背景要求较低,能适应较强的背景反光与阴影;避免对穗粒进行图像分割,使穗行线与行间隙定位的可靠性和准确性提高;校正单幅果穗图像的成像误差,使穗行数估算结果的准确性提高。统计结果显示:定位穗行线的失误率在6%以下,穗行数、行粒数和总籽粒数平均计算精度分别为98.84%、96.33%和95.67%,其中穗行数估算的零误差率为90.91%,平均测量速度达到每穗1.936 s。[结论]该方法能快速而精确地测量玉米果穗数量性状。[Objectives]The quantitative characteristics of maize ears are significant indexes for genetic traits measuring and production forecasting,including rows per ear,kernels per row and total kernels.Fast and accurate calculation for quantitative characteristics is a difficult problem of automatically measuring genetic traits of maize seed.[Methods]In this study,single image of maize ear was used as research object.An image acquisition device,which consisted of complementary metal oxide semiconductor(CMOS)camera,light emitting diode(LED)lights,light controller,stage and computer,was designed to capture images of maize ears.A maize ear was placed horizontally on the stage,and the inclination angle of ear was corrected by Radon formula after excessive image erosion.Based on the improved YCrCb color space,the contour of maize ear was obtained by bimodal method.Then the image was divided equally into 20 parts along the axis of ear.And the location points of ear rows in every part were obtained by pixel accumulation method.The lines of ear rows were obtained through linear interpolation of location points.Based on the changes of pixel values along the lines of ear rows,the kernels per row were accurately calculated.Through a geometric model of ear axis cross section,the rows per ear were accurately calculated.And an imaging error of single image was corrected by compensation method in order to improve its calculation accuracy.Finally,the total kernels were obtained by multiplying kernels per row with rows per ear.[Results]Experiment with dry corn demonstrated that the method,which was able to resist background reflections and shadows,had a low requirement to light environment and image background.And it avoided performing image segmentation for kernels of ear,thus the reliability and accuracy of positioning the ear rows’lines and gaps could be increased.And it corrected an imaging error of single image of maize ear,therefore the calculation accuracy of rows per ear could be increased.The experiment result demonstrated t

关 键 词：玉米果穗 数量性状 机器视觉 累计像素值分布 线性插值 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] S333.3[自动化与计算机技术—计算机科学与技术]