A field-based high-throughput method for acquiring canopy architecture using unmanned aerial vehicle images

Abstract

Plant architectural traits are important selection criteria in plant breeding that relate to photosynthetic efficiency and crop productivity. Conventional manual measures of architectural traits for large breeding trials are labour- and time-consuming. In this study, we proposed a new method to reconstruct three-dimensional (3D) canopy architectural models for high-throughput phenotyping of canopy architectural traits using image sequences acquired by an unmanned aerial vehicle (UAV) platform. The accuracy of UAV-derived models is evaluated by comparisons with models from 3D digitizing and measured values. The results indicated that the proposed method could obtain full canopy architecture in the early growth stages and the upper parts of the canopy architecture in the later growth stages. The leaf number, plant height, individual leaf area, and vertical and horizontal distributions of the leaf area estimated from UAV-derived models were in good agreements with the reference values for maize. The derived length and maximum width of individual leaves were close to the field measurements for maize (R2 > 0.92 for both, RMSE < 4.16 cm and 0.64 cm for blade length and maximum width, respectively) and soybean (R2 > 0.85 and RMSE < 0.77 cm). The newly proposed method has promising prospects for high-throughput phenotyping of the canopy architectural traits of field-grown crops and could facilitate the genotype selection in crop breeding and 3D plant modelling.