Presentation Information

[C19-02]3D phenotyping method for leaf morphology assuming anatomical structure

*Sogo Matsubara1, Koji Noshita1 (1. Kyushu University (Japan))

Keywords:

Morphometris,Plants,Plant Phenotyping

In plants, the leaf is an important organ for their survival. Leaf morphology is closely related to their functions, such as the photosynthetic rate and gas exchange efficiency. Therefore, plant phenotyping methods of leaf traits are important for understanding the relationship between morphology and function. Among the various leaf morphologies, the leaf contour structure is critical because it can be used to estimate leaf area and leaf angle. These characteristics are known to contribute to photosynthetic efficiency and crop yield. In this study, we developed a method for three-dimensional leaf tip and base reconstructions. First, the target is photographed from multiple viewpoints, and the two-dimensional coordinates of the tip and base of the leaf in each image are estimated using the deep neural network model, Keypoint R-CNN. The relative positions and orientations of cameras and the sparse point cloud of the plant are estimated from the multi-view images using Structure from Motion (SfM). Finally, 3D reconstruction is performed from the relative positions of the cameras and the 2D coordinates of the tip and base based on triangulation. This method has problems, such as incorrectly detecting the tip and base of a leaf in 2D images failing to estimate the occluded regions by other leaves. The proposed method contributes to accurate contour estimation in three dimensions and has potential applications in breeding and other areas.