玉米籽粒内部组分低对比度透射图像精确分割研究  

Accurate Segmentation of Low Contrast Transmission Images of Maize Grain Internal Components

作　　者：王金星[1,2] 李玉风 江浩 张宏建 张春庆[3] 刘双喜[1,4] WANG Jinxing;LI Yufeng;JIANG Hao;ZHANG Hongjian;ZHANG Chunqing;LIU Shuangxi(College of Mechanical and Electronic Engineering,Shandong Agricultural University,Taian 271018,China;Shandong Provincial Engineering Laboratory of Agricultural Equipment Intelligence,Taian 271018,China;College of Agronomy,Shandong Agricultural University,Taian 271018,China;Shandong Provincial Key Laboratory of Horticultural Machinery and Equipment,Taian 271018,China)

机构地区：[1]山东农业大学机械与电子工程学院,泰安271018 [2]山东省农业装备智能化工程实验室,泰安271018 [3]山东农业大学农学院,泰安271018 [4]山东省园艺机械与装备重点实验室,泰安271018

出　　处：《农业机械学报》2020年第6期238-247,共10页Transactions of the Chinese Society for Agricultural Machinery

基　　金：山东省自然科学基金面上项目(ZR201709180195);国家公益性行业农业科研专项(201303005);“双一流”奖补资金项目(SYL2017XTTD14)。

摘　　要：为解决玉米籽粒透射图像由于对比度较低造成内部组分提取不精确的问题,提出一种基于色彩通道非线性变换的多通道重叠区域分割方法,对玉米籽粒图像在灰度、R通道及b通道下得到的二值图像使用重叠区域原理,实现玉米籽粒组分的精确分割。首先,采集不同玉米品种的籽粒透射图像,提取单粒玉米籽粒;其次,采用多通道重叠区域法分割单粒籽粒图像,得到玉米籽粒胚部、角质胚乳和粉质胚乳3部分的图像;最后,以查全率和查准率评价多通道重叠区域法与传统图像分割方法对不同品种玉米籽粒的分割效果。不同品种玉米籽粒的分割试验表明:多通道重叠区域分割方法的查全率、查准率及综合评价指标均达到98%以上,分割效果优于传统的图像分割方法,能够实现不同品种透明角质玉米籽粒透射图像的精确组分分割。In order to solve the problem of inaccurate extraction of internal components in maize grain transmission image due to low contrast,a multi-channel overlapping region segmentation method based on non-linear transformation of color channel was proposed.The horny endosperm pixel information was obtained from the binary images of gray scale and R channel of maize;and the“farinaceous endosperm+embryo”image was transformed into Lab color space,the farinaceous endosperm image under b channel was extracted;and then the embryo pixel information was obtained from the binary images of R channel and b channel,so as to segment the components of maize grain accurately.Firstly,the transmission images of different maize varieties were collected,and the single grain was extracted.Secondly,the single maize grain images were segmented by the multi-channel overlapping region method based on the non-linear transformation of color channel,and the images of horny endosperm,farinaceous endosperm and embryo of maize grain were obtained.Finally,the recall ratio and the precision ratio were used to evaluate the segmentation effect of multi-channel overlapping region method and traditional image segmentation methods on different varieties of maize seeds.The experiments showed that the recall ratio,precision ratio and comprehensive evaluation index of the multi-channel overlapping region segmentation method were all over 98%.The segmentation effect was better than that of the traditional segmentation methods.It can realize the accurate segmentation of the transmission images of different varieties maize grain with transparent horny,and provide a scientific basis for the study of the three-dimensional distribution of internal components and the mechanism of dry crack formation in maize grains.

关 键 词：玉米籽粒 低对比度透射图像 组分分割 颜色空间变换 多通道重叠区域 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]