基于计算机视觉的玉米穗粒重无损测量方法  

Methods of Nondestructive Measurement for Kernel Weight of Maize Based on Multi-source Image Fusion

作　　者：周金辉[1] 马钦[1] 朱德海[1] 王越[1] 张晓东[1] 刘哲[1] 李绍明[1] 

机构地区：[1]中国农业大学信息与电气工程学院,北京100083

出　　处：《农机化研究》2015年第12期222-226,231,共6页Journal of Agricultural Mechanization Research

基　　金：中央高校基本科研业务费专项(2014JD042);公益行业(农业)科研专项(201203026);教育部博士点基金项目(20120008120015)

摘　　要：目前,国内玉米穗部考种大多采用目测尺量等传统手段,并且进行穗粒重测试时必须要经过脱粒环节,造成了对玉米果穗样本的破坏,无法保留考种原材料,不利于果穗精细化育种。为此,基于计算机视觉技术,利用可见光成像技术和X光成像技术,通过测量与计算果穗的三维几何性状,根据果穗质量分布特征建立了3种穗粒重估算模型:基于单粒重的穗粒重计算模型、基于容重的穗粒重计算模型和基于轴重的穗粒重计算模型。实验结果表明:基于单粒重的穗粒重计算模型和基于容重的穗粒重计算模型对果穗品种依赖较强,计算误差较强;相比较之下,基于单粒重的穗粒重计算模型比基于容重的计算模型拟合效果要更加好一些;相比于单粒重模型和容重模型,通过轴重模型来计算穗粒重有着较高的精确度,平均误差精度能达到1.30%,而且数据波动小,测量结果较稳定。该方法能够满足玉米果穗考种的基本需求,为实现考种流水化、智能化和自动化提供了很有价值的借鉴。At present , traditional methods such as eye-measurement and dipstick metering are generally used in maize varieties test interiorly , which is a waste of time and energy , and always account to a large calculation error .Besides , it has to be threshed before kernel weight testing , which means the tested materials could not be traced back , that is not suitable for extensive and meticulous breeding .This paper presents methods for extraction of 3 D phenotypic characteris-tics （ row number , bald length , grain plumpness and so on ） and internal structure features （ ear diameter , grain depth and so on ） of maize respectively based on visible light imaging technology and X optical transmission imaging technology . Furthermore , according to the parameters measured by these methods , three calculation models were built to calculate grain weight per ear： model based on single kernel weight , model based on volume-weight and model based on axis weight .The results show that error fluctuations of both models based on single kernel weight and based on volume -weight are relatively large and greater dependence on varieties , while the former is better than the latter .By comparison , the model based on single kernel weight is better .Nevertheless , the model based on cob weight has a high degree of accura-cy, whose average error precision can reach 1.30%, which has little data fluctuation , and the measurement results are stable , which meets the needs of extensive maize testing , and provides the basic data and model for the goal of high throughput and meticulous breeding .

关 键 词：穗粒重 无损测量 计算机视觉 可见光成像 X光成像 

分 类 号：S126[农业科学—农业基础科学]