Abstract: The automatic detection technology for plant phenotype plays an important role in agricultural reaserch and crop breeding. However, many 3D features cannot be extracted by 2D technology. Plant architecture (PA), including number of tillers, tiller angle and stem diameter, significantly affects the crop yield for many tillering crops. To acquire these characteristics parameters, traditional method needs huge manual labors, time-consuming, subjective and inaccurate. Therefore, it is impractical to perform manual phenotypic analysis. In order to automate PA parameters collection in crop breeding, a fast 3D reconstruction method was proposed to acquire high throughput PA characteristics parameters. To improve the reconstruction efficiency, parallel computing technique was used on a graphics processing unit (GPU). The processing time was approxemately 10 s of per plant on the Compute Unified Device Architecture (CUDA). This methodology was suitable for a high-throughput phenotyp testing platform.

Key words: 3D, volumetric reconstruction, plant architecture, GPU, high throughput

摘要： 植物表型自动化检测技术在农业研究和作物育种的过程中发挥了重要作用，但目前受限于二维技术三维特征很难被提取。株型是影响多分蘖作物产量的重要表型特征之一，它包括分蘖数、分蘖角、和茎粗等参数。传统方法中获取这些特征参数需要大量的人工测量，而人工测量具有耗时，主观性强，不准确等缺陷，因此用人工的方法进行大批量的表型分析是不现实的。为了使作物育种研究中株型参数提取实现自动化，提出一种用于高通量植株株型性状参数获取的快速三维重建方法，为了提高重建效率，研究中使用了图形处理单元(GPU)并行处理技术，在统一计算设备架构(CUDA)下进行重建的并行计算，使单株重建时间缩减到10秒左右，适合使用于高通量表型检测平台。

关键词: 3D, 体素重建, 株型, GPU, 高通量