S18 - Session P1 - Automatic flowering and apple detection in orchard with computer vision and machine learning

Authors: David Rousseau *, Mouad Zine El Abidine, Ali Ahmad

Computer vision and artificial intelligence promise to revolution horticulture with automation and more objective measurements of traits of agronomical importance. While many algorithms are already published in this domain [1] the transferability of this literature to the techno providers or the final users is often limited due to absence of sharing of the software or of the data set used to train these softwares. Some major initiatives to address the lack or reproducibility, in the plant domain at large, have so far been limited to plant model (Arabidopsis [2]) or major crops (Wheat [3]). Horticulture is therefore waiting for more similar initiatives. In this work, we present an annotated data set on apple detection and apple flowering that we make publicly available. This data set produced in collaboration with a group of European variety testing offices [4] on several sites is associated with base line algorithms (deep learning) already providing reasonable results. The challenge carried out by these data set are typical of orchard environment with background creating a major clutter with the targeted foreground. The test of new algorithms is made accessible via the deployment of a data challenge related to this data set. We will present the result of the data challenge which is open during the academic year 2021-2022 to master students in data science. https://www.quantitative-plant.org/ https://competitions.codalab.org/competitions/18405 http://www.global-wheat.com/2020-challenge/ https://www.h2020-invite.eu/