• Colombeau, Gallian
• Chapuis, Romain
• Thomas, Samuel
• David, Eric
• Negre, Vincent
• Weiss, Marie
• Adam, Boris
• Tireau, Anne
• Burger, Philippe
• Tong, Anthony
• Neveu, Pascal
• Pansanel, Jérôme
• Moreau, Patrick
• Baret, Frédéric

Abstract

High throughput phenotyping consists of measuring the characteristics of plants on different time and organisation scales. The PHENOME-EMPHASIS [1] project, associating INRAE, Arvalis and Terres-Inovia institutes, aims to develop broadband phenotyping infrastructures at national level. A particular effort has been made in recent years on the development of data acquisition systems (in particular drone and phenomobile) allowing the use of different types of sensors (high-resolution RGB, multispectral and thermal infrared cameras, LIDARs). All of these sensors generate a large volume of images, which must be processed, stored and archived.Prototype processing modules have been created within the UMT CAPTE [2] responsible for the development of high-speed field phenotyping methods and tools. They have been industrialised and integrated into the 4P data processing platform (Plant Phenotyping Processing Platform). Some modules incorporate proprietary software (Matlab, Photoscan), others require specific software libraries and can be run on heterogeneous environments (Windows, Linux). Portability is therefore a crucial element which is ensured by the encapsulation of the modules in Docker containers. These modules can be chained into workflows that users create according to their needs, using interactive tools available from a web interface developed in Java. A distributed computing architecture has been set up to execute these workflows: it is based on the Cromwell processing engine [3] which is responsible for executing the various modules sequentially or in parallel. Cromwell takes care of the sequencing of the modules but does not allow the processing to be distributed to several machines. The Docker Swarm tool allows the execution of Docker containers to be distributed to several machines in a cluster. This mode is based on one or more managers who will centralise the requests and distribute them to different nodes (machines on which the Docker service is installed). As far as data persistence is concerned, the raw data and processing outputs are stored on a distributed architecture based on iRODS technology [4]. The 4P platform is connected to the PHIS (Phenotyping Hybrid Information System) information system [5]. The objective of this information system is to store and organise the data produced within the framework of the PHENOME-EMPHASIS project by applying the FAIR (Findable, Accessible, Interoperable and Reusable) principles. This connection is made dynamically through the use of web services. The 4P platform is fully integrated into the France Grilles infrastructure [6]. It is a distributed infrastructure for the calculation and storage of scientific data providing various services to users, in particular : - the FG-CLOUD cloud service that allows the deployment of virtual machines on demand ; - the FG-IRODS service, which provides highly available and customisable storage; - the FG-DIRAC service, which enables grid computing.For the deployment of the 4P platform, we relied on the FG-CLOUD service for the application part and the FG-IRODS service for the persistent data storage part.The poster will detail the functionalities offered by the 4P platform, the technologies used and the technical infrastructure, in particular the integration with PHIS and France Grilles.Réeerences :[1] https://www.phenome-emphasis.fr/ [2] https://www6.paca.inra.fr/emmah/Programme-scientifique-et-Equipes/Equipe-CAPTE [3] https://github.com/broadinstitute/cromwell [4] https://irods.org/ [5] http://www.phis.inra.fr [6] http://www.france-grilles.frTranslated with www.DeepL.com/Translator (free version)

Topics

• impedance spectroscopy
• cluster