Table_1_Automatic Identification of First-Order Veins and Corolla Contours in Three-Dimensional Floral Images.docx

Defining and quantifying corolla traits are essential for studying corolla shape variation. Three-dimensional (3D) images of corollas contain comprehensive information regarding corolla structures and are optimal for studying corolla shapes. Conventionally, corolla traits are identified and quantified manually from 3D images. Manual identification is time consuming and labor intensive. In this study, approaches are proposed to automatically identify first-order veins and corolla contours in 3D corolla images. The first-order veins of the corollas were identified using Hessian of Gaussian and Dijkstra’s algorithm. The contours of the corollas were identified using vector harmony and node distance thresholding. A total of 130 3D images of 28 species in the subtribe Ligeriinae were collected and used to test the proposed approaches. The successful detection rate reached 86.54%. Two derived traits, contour–vein ratio and corolla angle, were defined and quantified using the first-order veins and corolla contour results to investigate the relationship between corolla shapes and pollination types of the subtribe Ligeriinae. Analyses revealed that the mean corolla contour, mean absolute corolla angle, and mean contour–vein ratio of the ornithophilic species were significantly smaller compared with the other species. The mean corolla contour, mean corolla angle, and mean contour–vein ratio of the melittophilic species were significantly larger compared with those of the ornithophilic species. The proposed method was also applied to certain Gesneriaceae species in the subtribes Gloxiniinae, Streptocarpinae, and Didymocarpinae. The results revealed that the method could be applied to most fresh sympetalous flowers for identifying first-order veins and corolla contours.

花冠性状的界定与量化，是开展花冠形态变异研究的必要前提。花冠的三维（3D）图像蕴含花冠结构的全面信息，是研究花冠形态的最优载体。传统研究中，研究者需手动从3D花冠图像中识别并量化花冠性状，但该方式耗时耗力。本研究提出一种可自动识别3D花冠图像中一级叶脉与花冠轮廓的方法：花冠一级叶脉的识别采用高斯黑塞矩阵（Hessian of Gaussian）与迪杰斯特拉算法（Dijkstra’s algorithm）实现；花冠轮廓的识别则通过向量调和与节点距离阈值法完成。本研究共收集了李戈里亚亚族（Ligeriinae）28个物种的130张3D花冠图像，用于验证所提方法的有效性，最终检测成功率可达86.54%。研究人员基于一级叶脉与花冠轮廓的识别结果，定义并量化了轮廓-叶脉比与花冠角度两个衍生性状，以探究李戈里亚亚族的花冠形态与传粉类型之间的关联。分析结果显示，嗜鸟媒物种的平均花冠轮廓、平均绝对花冠角度与平均轮廓-叶脉比均显著低于其他物种；而嗜蜂媒物种的上述三项指标则显著高于嗜鸟媒物种。本研究还将所提方法应用于苦苣苔科（Gesneriaceae）大岩桐亚族（Gloxiniinae）、旋果苣亚族（Streptocarpinae）与唇柱苣苔亚族（Didymocarpinae）的部分物种，结果表明该方法可适用于大多数新鲜合瓣花的一级叶脉与花冠轮廓识别任务。