Lower cost computer vision and machine learning

(Pejman Rasti, Etienne Belin, Natalia Sapoukina, David Rousseau)

This thematic is in the continuation of the current scientific project. Extension will be done in terms of observation scales and environment. While the main focus was the scale of the plant, observations will now also include the microscopic scale in collaboration with the microscopic platform of IRHS (IMAC). This is in accordance with the skills of the team developed in microscopic image processing in he EU project PROCHIP (2019-2022). In vitro imaging will be started for a variety of open questions including the monitoring of pathosystesm on agar gel or foliar disks, characterization of khal, nematode counting in variety testing. The group will address all the challenges associated with this imaging like the lighting conditions, dealing with condensation, distortion of agar gel, … This topic has been initiated within SNES in AKER Project (2012-2020) and  INRAE Colmar in SPE (2019-2020) project. Outdoor computer vision will be developed for the characterization of trees (3D shape, fruit development, biotic and abiotic stress detection) in the orchard of INRAE. This will be developed in collaboration with UE Horti in Angers on apple trees, within the EU project INVITE (2019-2024) or possibly on rosebush with GDO (IRHS).

Simulation assisted phenotyping

(Angélina El Ghazeri, Etienne Belin, Gehard Buck Sorlin, David Rousseau)

In March 2020, PHENOTIC is equipped with a new robot (PHENOBEAN) capable of monitoring the growth of plant with controlled light in terms of directivity, wavelength and intensity. The system is equipped with a robotic arm bringing a sensor in any position of the room. In parallel with experiments carried by the platform, ImHorPhen proposes to simulate Phenobean with ecophysiological models in order to optimize the experimental parameters which impact the growth of plants. This new thematic will be supported by a PHD to start in sept. 2020.  The simulator will be used to optimized growth but also to constitute automatically annotated models to serve as training data set for computer vision. An automatic control approach of the growth of plant will also be investigated with the team working on automatic control lab at LARIS (UA). This will be done in collaboration with STRAGENE team at IRHS.

Machine learning based data mining

(Angélina El Ghazeri, Julie Bourbeillon, Nizar Bouhlel, David Rousseau)

This new thematic will address the challenges of the flow of data created via the PHENOTIC platform. A typical run of phenotyping will produce some Terabytes of data. While images start to be processed efficiently via the first thematic, it is now necessary to go beyond and produce knowledge for the data extracted from these images. Like the design of computer vision algorithm, data mining could also benefit from the help of machine learning for automatic design of descriptive, predictive diagnostic analytics and prescriptive analytics. Such problematics will be investigated for data at rest and during the phenotying run.