Abstract Detail

The Potential of Machine Learning for Plant Biology

Bonnet, Pierre [1], Adan, Mora-Fallas [2], Mata-Montero, Erick [3], Hervé, Goëau [1], Alexis, Joly [4].

Challenges of plant species detection and identification by autonomous robot.

Plant species identification based on automated analysis of visual contents has drastically progressed in the last ten years (Wäldchen et al., 2018) . The plant identification challenge of CLEF (PlantCLEF), which runs since 2011 as part of the LifeCLEF lab (Joly et al., 2018), offers in particular seven-year follow-up of the progress made in this domain. The difficulty of the task and the volume of data were considerably enriched along the years. More precisely, the number of species was increased from 71 species in 2011 to 10,000 species in 2018 (illustrated by more than 1 million images). This continuous scaling-up was made possible thanks to the close collaboration with several important actors in the digital botany domain.
As illustrated in Bonnet et al. (2018), the best approaches are now very close to the human expertise, for taxa that can be illustrated with an appropriate volume of training data. These results open new opportunities to evaluate capacity by autonomous robots to make automated plant species identification. In this context, there is no intentional action of the photographer (in this case a robot), to take a picture of plant in front of him, as is the case with smartphone users who are able to take close up of the individual plant that they want to identify. Allowing species identification by autonomous robots, needs to develop a computational framework that will automatically detect and identify plant species in a visual data stream that will be much more noisy than plant images produced by humans.
We propose to present results that we obtained on weed identification from camera embedded on agricultural solar robot, used for weed removal. We will discuss efficiency and limits of such approach, based on Mask-RCNN method, in order to highlight new opportunities offer by deep learning technologies in plant sciences.
Wäldchen, J., Rzanny, M., Seeland, M., & Mäder, P. (2018). Automated plant species identification—Trends and future directions. PLoS computational biology, 14(4), e1005993.
Joly, A., ... & Müller, H. (2018). Overview of lifeclef 2018: a large-scale evaluation of species identification and recommendation algorithms in the era of ai. In International Conference of the Cross-Language Evaluation Forum for European Languages (pp. 247-266). Springer, Cham.Bonnet, P., .. &
Joly, A. (2018). Plant identification: experts vs. machines in the era of deep learning. In Multimedia Tools and Applications for Environmental & Biodiversity Informatics (pp. 131-149). Springer, Cham.

1 - CIRAD, BIOS, Cirad Umr Amap - Ta A-51/ps1, Boulevard De La Lironde, Montpellier Cedex 5 (France), 34398, France
2 - Costa Rica Institute of Technology, School of Computing, Cartago Central Campus Technology, Cartago, Costa Rica
3 - Costa Rica Institute of Technology, Computing, Calle 15, Avenida 14, 1 km Sur de la Basílica de los Ángeles, Cartago, Cartago, 30101, Costa Rica
4 - Inria, Zenith Team, 161 rue Ada, Montpellier, 34095, France

deep learning
Instance detection
plant identification
Plant detection
Weed control
agricultural robot.

Presentation Type: Symposium Presentation
Abstract ID:804
Candidate for Awards:None

Copyright © 2000-2019, Botanical Society of America. All rights reserved