基于多视图几何的白菜薹分割与关键表型测量  

Segmentation and measurement of key phenotype for Chinese cabbage sprout using multi-view geometry

作　　者：王瑞萍 刘东风 王先琳 杨会君[1,3,4] Wang Ruiping;Liu Dongfeng;Wang Xianlin;Yang Huijun(College of Information Engineering,Northwest A&F University,Yangling 712100,China;Shenzhen Agricultural Science and Technology Promotion Center,Nanshan 518057,China;Shaanxi Key Laboratory of Agricultural Information Perception and Intelligent Service,Yangling 712100,China;Key Laboratory of Agricultural Internet of Things,Ministry of Agriculture and Rural Affairs,Yangling 712100,China)

机构地区：[1]西北农林科技大学信息工程学院,杨凌712100 [2]深圳市农业科技促进中心,南山518057 [3]陕西省农业信息感知与智能服务重点实验室,杨凌712100 [4]农业农村部农业物联网重点实验室,杨凌712100

出　　处：《农业工程学报》2022年第16期243-251,共9页Transactions of the Chinese Society of Agricultural Engineering

基　　金：陕西省重点研发计划项目(2021NY-179);广东省乡村振兴战略专项-农业生产发展项目(2130122)。

摘　　要：植物表型调查是选育优良品种和基因功能研究的重要依据,为理解植物生长发育规律及环境的作用提供有力支持。针对传统叶菜类植物表型分析方法存在速度慢、误差大、维度限制等问题,该研究提出了一种基于高通量重建和茎叶自动分割的白菜薹关键表型参数提取方法。首先,基于多视图立体几何技术对白菜薹进行多视角RGB图像三维重建、尺度恢复、均匀简化、背景去除及点云去噪等预处理。之后,提出基于超体素的改进植物器官自动分割算法,将植株分为茎、叶片等不同语义类别。在此基础上,给出有效的表型参数计算方法,完成了株高、叶长、颜色等7个关键性状的无损、精确测量。试验结果表明,该研究实现了白菜薹关键表型自动分析,茎叶器官分割的精确率、召回率及F1分数的均值分别为0.961、0.940、0.943;株高、株幅、叶长、叶宽的均方根误差分别为0.261、0.313、0.174、0.100 cm,叶面积及叶片数的均方根误差分别为1.608 cm^(2)和0.283,平均绝对百分比误差分别为1.659%、1.643%、1.417%、2.486%、8.258%、6.000%。与其他方法相比,该研究具有较低的综合误差,可适应叶片形状不规则的植物表型参数提取研究。同时,克服了当前植物冠层幼叶难以分割、表型性状提取效率低等困难,为精准农业领域叶菜表型高效分析提供有效的技术手段,可在进一步的基因型到表型研究中发挥重要作用。Plant phenotype has been one of the most important indicators to select and breed superior varieties in modern agriculture.The traditional phenotypic analysis of leafy vegetables cannot fully meet the requirement of large-scale production in recent years,such as the slow speed,large error,and limited dimension.Moreover,most morphological measurements have been currently confined to only several plants(such as maize,cotton and sorghum),particularly with the complex procedure and low automation.Taking the Chinese cabbage sprout as the research object,this study aims to extract the key phenotypic parameters of the plant using the high-throughput reconstruction and automatic segmentation of stem and leaf.Firstly,a multi-view geometry was utilized to reconstruct the three-dimensional model of Chinese cabbage sprout from sequence images.A series of pre-processing operations were used to establish the three-dimensional model of pure plants with the actual scale,including scale recovery,background removal,point cloud denoising,and uniform simplification.Secondly,the stem and leaf organs were automatically segmented using the convexity criterion and random sampling consistency.Thirdly,the principal component analysis and directed bounding box were combined to measure the plant height and width for the phenotypic parameters.The number of leaves was counted by the cluster segmentation of leaf clusters.The shortest path searching was selected to accurately calculate the leaf length and width.A greedy projection triangulation was used to calculate the leaf area.An HSV model and mean color histogram were utilized to measure the color characteristics for distinguishing health status of plants.The classification accuracy,recall,and F1-score were 0.922,0.938,and 0.930,respectively.Finally,the segmentation experiments were carried out on the six varieties of Chinese cabbage sprouts at different growth stages.A comparison was then made with the ground truth.It was found that the parts belonging to stem and leaf were segmented corre

关 键 词：三维 数字化 茎叶分割 表型测量 白菜薹 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]