PhénaufolMise au point d'outils et techniques de PHENotypage pour détecter AUtomatiquement les maladies FOLiaires de la betterave.

Abstract

The Phenaufol project consisted in the development of a robotic phenotyping process for sugar beet trial plots, and of algorithms for automated quantification of symptoms. A preliminary go/no-go approach allowed us to focus on the most suitable sensors for leaf diseases phenotyping. In order to measure the impact of each main foliar disease, several image analysis methods (thresholding, texture, machine learning) were compared. At the same time, a mathematical modeling of the chosen robotic rig was done for a precise motion execution. Several in-field phenotyping campaigns were conducted to validate the system improvements and to collect disease dynamics data. Geographic visualization was done as a proof of concept. Finally, a vision / robotics pairing was implemented with a tracking algorithm, to move the camera at the closest to the symptoms. Some blockages still need to be removed before running a wide-scale phenotyping campaign (many commercial varieties, national network of experimental sites).

Code and data availability

The article describes the Phénaufol robotic phenotyping pipeline and image analysis methods but provides no public dataset, image collection, author code repository, or trained model. The only URLs cited (Keras, TensorFlow, Python) are generic third-party libraries, not paper-specific assets.

No evidence-backed public reproduction asset is currently recorded.