基于点云的谷粒高通量表型信息自动提取技术

在进行水稻的数字化考种、表型与基因关联分析和数字农业仿真模拟时,需要大量的谷粒表型信息作数据支撑。本文提出了一种基于三维点云的谷粒高通量表型信息自动提取方法,能同时自动获取谷粒的三维模型和40个表型参数,实现谷粒形状的定量和定性描述。首先,通过对谷粒点云数据进行聚类分析,完成谷粒点云的分类;其次,实现谷粒的三维重建,对谷粒离散点云进行柱面构网,获取谷粒点云的三维模型数据;最后,根据不同表型参数的特点,实现了谷粒的三维表面积和体积、长、宽、高、3个主成分剖面的周长和面积等11个基本参数与长宽比、长高比和体积比等11个衍生参数以及18个形状因子的自动提取。利用Handyscan 700型手持式激光扫描仪获取的谷粒高精度点云数据进行实验,成功实现了谷粒表型参数的自动提取,测量结果可达毫米级。基于主成分方法分析了各表型参数的权重。以游标卡尺测量值和Geomagic Studio测量值作为真值,长、宽、高的平均相对误差为1.14%、1.15%和1.62%,体积和表面积的相对误差为零,3个主成分剖面面积的平均相对误差为1.82%、2.12%和2.43%。本文方法与人工测量方法及软件测量方法相比,精度相当,且具有批量、自动、人工干预少(仅数据采集阶段需要人工操作)以及效率高的特点。

Automatic Extraction of High-throughput Phenotypic Information of Grain Based on Point Cloud

Large amount of grain phenotypic information is needed in researches such as digital grain traits investigation, phenotype and gene association analysis and digital agriculture simulation. A method for automatic extraction of grain high-throughput phenotypic information based on point cloud was proposed, aiming to automatically obtain three-dimensional (3D) grain model and 40 phenotypic parameters. Firstly, the classification of grain point cloud was completed through cluster analysis. Secondly, 3D grain model was reconstructed with cylindrical mesh method. Finally, according to the characteristics of different phenotypic parameters, 11 primary parameters, 11 derived parameters and 18 shape factors were automatically extracted. Experiment using data obtained by hand-held laser scanner (Handyscan 700) showed that the measurement result could reach millimeter level. The weight of each phenotypic parameter was analyzed based on principal component analysis method. With parameters measured by vernier caliper and Geomagic Studio as the true value, the average relative error of length, width, height, surface area and volume, the cross-sectional area of three principal component sections was 1.14%, 1.15%, 1.62%, 0, 1.82%, 2.12% and 2.43%, respectively. Compared with the manual measurement method and the software measurement method, the results of the proposed method was competitively accurate, which had advantages of batch processing, automation, less manual intervention (only in data acquisition) and high efficiency.