AUTOMATIC GENERATION OF GEOMETRIC PARAMETERS OF INDIVIDUAL CAULIFLOWER PLANTS FOR RAPID PHENOTYPING USING DRONE IMAGES

Abstract

Abstract. Multitemporal drone surveys are a perfect tool to determine various geometric and spectral crop parameters for rapid phenotyping in field trials. Depending on the geometric resolution and the size of the crop, information at leaf level or canopy level can be obtained. The focus of this paper is to demonstrate which geometric properties can be automatically derived from high resolution drone imagery during the vegetation period. For this research approx. 1920 cauliflower with a large genetic variety were planted and monitored by five different drone surveys at an altitude of 20 m, using a high resolution 36 Mpix. RGB-camera. In order to minimize intensive radiometric calibration, BRDF effects and eliminate shade, flights were carried out at overcast skies. After photogrammetric image processing, detailed crop height models (CHM) were computed. 10 distinct crop parameters were derived from a combination of the orthophotos, the CHM and additional information. According to the phenological phase a specific set of parameters was developed for every flight. For instance, the position of the individual plants is computed right after the first flight. For the flight prior to harvesting, an algorithm for the head diameter and the curvature of the cauliflower heads was developed. Geometric parameters are generally better suited for automation, because they require less specific ground truth or reference information, than spectrally derived biophysical parameters.

Plant phenotyping relevance

ドローン画像と写真測量から個体別の草高、位置、花球径、曲率などの植物形質を自動抽出する手法が研究の中心であり、圃場フェノタイピング手法に該当する。

Code and data availability

The paper describes drone-based cauliflower phenotyping (CHMs, head diameter, curvature, coverage) but contains no data availability statement, no public dataset or image deposit, and no code release. The only mentioned implementation is ArcGIS models with no availability language or URL. No paper-specific public asset

No evidence-backed public reproduction asset is currently recorded.