lexiconMaize tassel detection on drone RGB images
2025 · Arvalis
[Poster] This study presents a research project dedicated to the automatic detection of maize tassels from RGB images acquired by unmanned aerial vehicles (UAVs). The objective is to develop a high-throughput phenotyping pipeline enabling the identification, counting, and characterization of tassels, in order to improve the understanding of flowering and reproductive processes, as well as their relationship with yield and plant adaptation to environmental stresses. The proposed workflow includes: (i) standard pre-processing steps (image fusion, filtering, photogrammetry, and extraction of image subsets) carried out on the INRAE 4P platform; (ii) image pre-selection based on quality and row visibility using the Depth Anything model; and (iii) tassel detection through a deep learning approach (YOLOv8x), trained on a large annotated dataset comprising approximately 3,500 images and over 100,000 tassels collected from public sources and the DiaPhen platform. Results demonstrate robust detection and counting performance, showing strong correlation with manual observations, although with a slight underestimation in high-density areas or at early developmental stages
- Voir sur la plateforme R&D agricole · Document
- Projet : Programme transversal 2025 « Mobilisation du levier du numérique pour soutenir la conception, le pilotage, le déploiement et la valorisation de systèmes de production agricole innovants et performants »
Source : Plateforme R&D Agricole (ACTA) — CC BY-NC-SA 4.0. Usage non commercial uniquement.