High-throughput calculation of organ-scale traits with reconstructed accurate 3D canopy structures using a UAV RGB camera with an advanced cross-circling oblique route

Abstract

The measurement of organ-scale traits in large-scale fields remains a bottleneck in genotype-phenotype association studies for crop improvement. To address this issue, an advanced cross-circling oblique (CCO) route was proposed, consisting of multiple single-circle routes. Using multi-view images from lightweight UAVs with the CCO route, 3D crop canopies were reconstructed for maize, cotton, and sugar beet. Organ-scale traits were estimated from the CCO-derived and traditional five-directional oblique (FDO)-derived 3D canopy models and were compared to manual measurements. An algorithm was further proposed to compare the image utilisation efficiency between the CCO route and FDO route with three indicators, including the total utilisation (TU), effective utilisation (EU), and occlusion rate (OR). The results showed that the proposed CCO route can obtain complete canopy structures throughout the growth stages for crops with relatively wide-row planting, such as maize. For densely planted crops like cotton, the full canopy structure at early growth stages (47 d after sowing (DAS)) and the upper parts of the canopy architecture at later growth stages (71 and 90 DAS) could be obtained. The leaf length and width from different layers estimated from CCO-derived models were in good agreement with manual measurements for maize (R2 = 0.93, RMSE = 3.05 cm, nRMSE = 4.7% and R2 = 0.88, RMSE = 0.49 cm, nRMSE = 5.5% for leaf length and leaf width, respectively) and cotton (R2 = 0.81, RMSE = 1.16 cm, nRMSE = 8.6% and R2 = 0.93, RMSE = 0.99 cm, nRMSE = 7.2% for leaf length and leaf width, respectively). CCO-derived 3D canopy models also provided higher accuracy for organ-scale trait estimation than FDO-derived 3D canopy models (R2 of 0.80 versus 0.76 for both leaf length and width estimation, respectively). Regarding image utilisation, the CCO route outperformed the FDO route, with an average value of 124% higher in EU and 11.7% lower in OR, tested on sugar beet varieties with different canopy structures. The proposed CCO route adopts a non-linear mobile route to obtain canopy images from more perspectives, taking into account both the model accuracy and efficiency. Although the flight altitude of the CCO in this study was relatively low (4 m above the canopy), the same model accuracy can be achieved at higher flight altitudes using a higher-quality camera, thus significantly reducing image acquisition time. This study represents the first time that accurate crop structures in large-scale fields were characterised using a method featuring mobility, affordability, throughput, accuracy, and efficiency. The proposed method could provide novel opportunities for accelerating plant breeding and precision agriculture.