基于图像处理技术的菜用大豆豆荚高通量表型采集与分析  

High-Throughput Phenotype Collection and Analysis of Vegetable Soybean Pod Based on Image Processing Technology

作　　者：张小斌[1] 谢宝良 朱怡航 郑可锋[1] 顾清[1] ZHANG Xiaobin;XIE Baoliang;ZHU Yihang;ZHENG Kefeng;GU Qing(Institute of Digital Agriculture,Zhejiang Academy of Agricultural Sciences,Hangzhou,Zhejiang 310021)

机构地区：[1]浙江省农业科学院数字农业研究所,浙江杭州310021

出　　处：《核农学报》2022年第3期602-612,共11页Journal of Nuclear Agricultural Sciences

基　　金：浙江省重点研发计划项目(2021C02052)。

摘　　要：为实现菜用大豆豆荚表型的高通量采集与分析,本研究利用可见光成像技术获取豆荚图像,综合考虑育种工作对表型信息的需求,提出了一套基于计算机图像处理技术的菜用大豆豆荚表型信息采集分析方法,并开发了相应的分析系统软件。研究通过试验确定最优参数,实现了菜用大豆豆荚长度、宽度、弦长、弧长、面积、每荚豆粒数、弯曲度、标准色距等表型性状的自动化、批量化提取。通过标定物与豆荚实物的测量验证,图像分析结果可以达到与实际测量值相当的精确度。图像分析得出的豆荚表型指标与人工实际测量值无显著差异(P>0.05),测量误差值均小于0.07,决定系数R^(2)在0.95以上。用水平分割的方法测算每荚豆粒数时,豆荚图像的最优分割数量为15。此外,从豆荚整体的弯曲程度出发,利用豆荚中心点连接上下端点形成的夹角作为衡量豆荚弯曲度的指标,对豆荚弯曲度的描述更为准确与合理。豆荚颜色与标准颜色参考值之间的标准色距能够实现对豆荚颜色性状与育种目标之间差距的定量化评估。本研究提出的菜用大豆豆荚表型采集技术操作简便、成本较低,能减少人工测量误差,提高表型信息可靠性,大幅提升育种工作的效率;同时,此项技术有助于毛豆表型信息的定量化分析和标准化表型数据库的建立。In order to realize the high-throughput acquisition of vegetable soybean pod phenotypes,a visible light imaging technology was used to obtain bean-pod images.Considering the requirements of breeding for phenotypic information,the acquisition and analysis methods for the soybean pod phenotypic information was proposed based on a computer image processing technology,and corresponding analysis software was developed.The proposed method can be used to automatically extract the phenotypic information from vegetable soybean pod images such as pod length,pod width,chord length,arc length,pod area,number of seeds per pod,bending degree and standard color distance in large quantities.The result of image analysis was verified by the measurement of calibration objects and soybean pods.There was no significant difference between the phenotypic indexes of pods obtained from the images and those measured manually(P>0.05),with measurement errors less than 0.07 and determination coefficients(R^(2))higher than 0.95.The horizontal segmentation method was used to calculate the number of seeds per pod,and the optimal segmentation number was 15.This study proposed a method to measure the pod bending degree using the angle formed by the line segments connecting the central point of the pod with the upper and lower endpoints of the pod respectively,which is more accurate and reasonable.The standard color distance between the pod color and a standard color reference can be used to quantitatively evaluate the difference between pod color traits and breeding objectives.The proposed phenotype acquisition technology is simple and low-cost.It can reduce the artificial measurement errors,improve the reliability of phenotypic information,and greatly increase the efficiency of breeding work.Additionally,this technique is helpful for the quantitative analysis of vegetable soybean phenotypic information and the establishment of a standardized phenotype database.

关 键 词：菜用大豆 辅助育种 表型 图像识别 高通量 

分 类 号：S643.7[农业科学—蔬菜学]