种子动态萌发表型分析算法的研究和软件实现  

Developing vision-based analytic algorithms and software to dynamically measure key traits in seed germination

作　　者：赵建华[1,2] 周洁 戴杰 丁国辉 徐凌翔 关雪莹 周济 ZHAO Jianhua;ZHOU Jie;DAI Jie;DING Guohui;XU Lingxiang;GUAN Xueying;ZHOU Ji(Academy for Advanced Interdisciplinary Studies/Plant Phenomics Research Center,Nanjing Agricultural University,Nanjing 210095,China;College of Artificial Intelligence,Nanjing Agricultural University,Nanjing 210031,China;College of Agriculture and Biotechnology,Zhejiang University,Hangzhou 310029,China;National Institute of Agricultural Botany/Cambridge Crop Science,Cambridge CB30LE,The United Kingdom)

机构地区：[1]南京农业大学前沿交叉研究院/作物表型组学交叉研究中心,江苏南京210095 [2]南京农业大学人工智能学院,江苏南京210031 [3]浙江大学农业与生物技术学院,浙江杭州310029 [4]英国国立农业植物研究所/英国剑桥作物研究中心,剑桥CB30LE,英国

出　　处：《南京农业大学学报》2022年第6期1266-1275,共10页Journal of Nanjing Agricultural University

基　　金：江苏省基础研究计划项目(BK20191311);中央高校基本科研业务费专项资金(JCQY201902)。

摘　　要：[目的]种子是植物研究中重要的对象之一,本研究旨在对关键种子萌发表型进行动态监测及量化分析,为了解不同植物生存、生长和繁衍提供表型依据。[方法]本研究以小麦(Triticum aestivum)为研究对象,利用监督式机器学习算法(如K近邻、支持向量机、随机森林),在不同颜色空间上对3种弱筋小麦品种的种子萌发图像序列进行前、背景对象训练及背景分割,然后通过构建自动化图像处理算法进行目标提取,再结合图论和二维骨架动态分析幼根和根尖点的位置变化,实现关键萌发性状的高通量数字化提取。[结果]本研究可获得大量人工难以计量的萌发性状,包括种子长、宽、面积、周长,幼根和幼芽长度及生长速率等。通过与人工统计数据的线性回归分析,关键动态性状如幼根长、根生长速率、芽长的决定系数R 2值分别为0.922(n=188,P<0.001,RMSE=1.727)、0.719(n=191,P<0.001,RMSE=0.406)、0.897(n=115,P<0.001,RMSE=2.726)。[结论]本研究提出的算法能有效获取种子萌发动态表型组,并可扩展至棉花(Gossypium barbadense)和油菜(Brassica napus),为遗传育种和植物研究提供表型分型依据和智能化解析技术。[Objectives]Seed is one of the most important research topics in plant research.Our research aimed to dynamically measure key germination-related traits,so that important phenotypic evidence can be provided for advancing our understanding of plant survival,growth,development,and reproduction.[Methods]Utilizing wheat(Triticum aestivum)as a model plant,we used automated image analysis together with machine learning algorithms(e.g.K-Nearest Neighbors,Support Vector Machine,Random Forests)to train foreground and background objects,followed by background segmentation and object extraction based on image series collected from three weak gluten wheat varieties.Then,graph theory and two-dimensional skeletonization were employed to dynamically analyze changes of radicles and radicle tip positions to measure key germination-related traits in a high-throughput manner.[Results]We had collected a range of phenotypic traits in this study that were difficult to obtain through traditional approaches,including seed length,width,area,perimeter,radicle and seedling length,and their growth rates.We applied a linear regression analysis to validate the computational analysis results against manual scoring.The square of the correlation coefficient(R 2)was computed for traits such as radicle length,radicle growth rate and seedling length;these are 0.922(n=188,P<0.001,RMSE=1.727),0.719(n=191,P<0.001,RMSE=0.406),0.897(n=115,P<0.001,RMSE=2.726),respectively.[Conclusions]The results suggest that the algorithm and open-source software presented here can reliably obtain dynamic seed germination traits,which can also be extended to other crop species such as cotton(Gossypium barbadense)and oilseed rape(Brassica napus),providing phenotypic evidence and smart analytic solutions to enable studies in plant genetics and crop breeding.

关 键 词：种子萌发 动态表型分析 自动化图像处理 监督式机器学习 小麦 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]