Three-dimensional image-based modelling of linear features for plant biomass estimation

Abstract

Biomass estimation is important for biological research and agricultural management. Low-cost two-dimensional (2D) computer vision has been applied to non-contact biomass estimation. However, the rapid increase of computing power has enabled the use of stereo vision models for this purpose. The objectives of this study were to develop an alternative 3D image-based reconstruction model that utilizes geometric features of plant leaves for estimation of biomass and to evaluate its robustness to varying illumination conditions, complex plant geometry, and leaf surface texture. At its core, the proposed model extracts and matches linear features that are then reconstructed in 3D space. As linear features characterize the plant's silhouette and entities on its surface, a detailed and accurate 3D reconstruction of its shape can be obtained. The algorithm performance was evaluated both in greenhouse and field studies, showing accurate estimation under varying canopy geometries, growth stages, and illumination conditions. Results show an ability to accurately estimate height (error ˜4.5%) and leaf cover area (error ˜4.5%) values under these conditions. Additionally, a strong linear relation was obtained between estimated plant volume and measured biomass (R 2 ˜ 0.92), which proved to be an accurate predictor when applied on new plants (error ˜4.5%). These abilities make this model a promising tool for future development of biological models and precision agricultural practices.