Crop height estimation based on UAV images: Methods, errors, and strategies

Abstract

Unmanned aerial vehicles (UAVs) have emerged as a promising platform for determining the dynamic phenotypic traits of crops in the field in a rapid and cost-effective manner. Crop height is a common and important phenotypic trait, and its acquisition with high accuracy usually requires spatial auxiliary (SA) information, such as a digital terrain model in the early growing season, digital surface models later in the season, ground control points, and ground truth of crop height. The reasonable selection of SA information involves balancing the cost and accuracy of crop height acquisition, but this problem has not been systematically studied and it needs to be resolved urgently in the agricultural industry. In this study, we compared four rapeseed height estimation methods using UAV images collected at three growth stages based on the structure from motion algorithm, where one method had complete data and the other three had incomplete SA information. To reduce the crop height estimation errors with incomplete data, improved methods were developed to construct the missing SA information. The optimum results were obtained using complete SA information, where R² was 0.932 and the root mean square error (RMSE) was 0.026 m. For crop height acquisition using incomplete data, the R² values could be controlled above 0.445 and RMSE below 0.146 m. In this study, systematic strategies were developed for selecting appropriate methods to acquire crop height with reasonable accuracy while balancing the cost requirement for use in scientific research and agricultural production.

Plant phenotyping relevance

UAV画像とSfMを用いた作物高という植物形質の取得法を比較・改良し、誤差と精度を評価しているため、フェノタイピング手法が研究の中心です。

Code and data availability

公開本文の所在を確認できませんでした。非公開または購読が必要な可能性があります。

No evidence-backed public reproduction asset is currently recorded.