植株自旋转多视角重建技术在小麦植株三维表型获取中的应用评估

基于多视角重建技术的作物三维表型高通量获取系统成本低、获取效率高，引起越来越多的关注。植物自旋转式拍摄平台易于搭建，但植物旋转过程中产生的抖动对点云三维重建和表型解析精度有一定影响。为评估旋转式多视角成像在小麦植株三维表型解析中的适用性，基于植物旋转设计了便携式小麦植株三维表型高通量采集系统，选取穗期不同品种的小麦植株作为实验样本进行点云重建，基于Hausdorff距离评价了重建点云的精度误差；并基于人工测量数据，对所提取的表型指标精度进行评价。结果表明，植物旋转式重建的点云与相机旋转式重建的点云有较高的一致性，点云精度差距基本控制在0.4 cm以下；获取的叶长、叶宽和株高的均根方误差分别为0.79、0.13和0.53 cm，平均绝对百分比误差分别为3.26%、7.63%和0.74%，表明该方式适合穗期的小麦植株表型重建，具有较高的点云重建和表型提取精度，并为小麦植株表型评价提供了一种低成本的解决方案。

Evaluation of Plant Self-rotation Multi-view Reconstruction Technique in 3D Phenotype Acquisition of Wheat Plants

The high-throughput acquisition system of crop phenotypes based on multi-view 3D reconstruction technology with low cost and high acquisition efficiency has attracted more and more attentions. The plant self-rotation capture platform is easy to build and low cost， but the jitter generated during plant rotation has an impact on the point cloud 3D reconstruction and phenotype resolution accuracy. To evaluate the applicability of plant rotational multi-view imaging in 3D phenotype resolution of wheat plants， designed a portable 3D phenotype high-throughput acquisition system for wheat plants based on plant self-rotation， and wheat plants of different varieties at the spike stage were selected as experimental samples for point cloud reconstruction， and the accuracy error of the reconstructed point cloud was evaluated based on Hausdorff distance； the accuracy of the extracted phenotypic indicators was evaluated based on manual measurement data. The experimental results showed that the point clouds reconstructed by plant self-rotation had high consistency with those reconstructed by camera rotation， and the difference of point cloud accuracy was basically controlled below 0.4 cm； the root mean square errors （RMSE） of obtained leaf length， leaf width and plant height were 0.79， 0.13 and 0.53 cm， and the mean absolute percent errors （MAPE） were 3.26%， 7.63% and 0.74%， respectively， indicating that this approach was suitable for phenotypic reconstruction of wheat plants at the spike stage and had high accuracy of point cloud reconstruction and phenotype extraction， which provided a low-cost solution for wheat plant phenotyping.