#DigitAg Master2 Internship Offers 2022

Every year, #DigitAg grants Master 2 Internships for French and foreign students

In 2022, subjects from different disciplines are suggested in Human&Economic Sciences, Environment&Life sciences, as well as in Engineering and Mathematics.

  • To apply please send your curriculum and application letter to the relevant contact from the list below
  • Allowance: #DigitAg gives internships grants to the research units:  the amount of your allowance is to be asked to the contact person

Maths and its applications

Environnement and life sciences

Human and Economic sciences

Engineering sciences

Maths and its applications

Improvement of an event detection tool (reproductive, sanitary, malfunction) in a herd of dairy cows and several groups of sows

Keywords : precision farming, event detection, water consumption, multi-species

  • Duration (in months): 6
  • Desired start date: 1/02/2022
  • Research unit: Pegase, Inrae
  • Contact: Charlotte GAILLARD – charlotte.gaillard[AT]inrae.fr
  • #DigitAg: Axis 5, Challenge 2

Precision farming tools combined with the breeder’s observations, allow individual and automated tracking of dairy cows and sows contributing to an earlier detection of events (calving, health problems, malfunction). From water consumption data, an analysis method made it possible to detect disturbances related to health, reproduction or technical dysfunction events. This method is more than 95% specific for cows and sows, however its sensitivity is at best around 70% for cows and remains lower for sows (<50%).
The objective of this internship is to improve this method of identifying disturbances. The main hypothesis of this internship is that certain variables are dependent on each other (eg amount of water consumption and amount of feed ingested for cows) and that the structure of dependence between these variables will change during a disturbance event. The combined study of the dependent variables should make it possible to improve the sensitivity of our method of identifying disturbances.
This internship will be carried out at INRAE UMR PEGASE, Saint-Gilles, in collaboration with UMR GenPhySE from INRAE and will be co-supervised by a nutritionist researcher in cattle breeding (Anne Boudon, PEGASE), a nutritionist-modeler researcher in pig breeding (Charlotte Gaillard, PEGASE) and a statistician researcher (Tom Rohmer, GenPhySE). A follow-up committee of the internship will bring together researchers competent in the physiological regulations of water consumption and in the processing of dynamic data in livestock.
At the beginning of the internship, the databases will be available as well as a first version of an event detection program based on a differential smoothing method. The first step of the internship will consist in determining pairs of dependent variables specific to each species. The second step will consist of improving the event detection process by applying it to these pairs.

Supporting decision-making in precision spraying with a study on the relationship between statistical distribution and 3D spatial covering of plant protection products in plants
Filled internship

Keywords : geostatistics, 3D spatial data, grapevine, spraying, LiDAR, precision agriculture

  • Duration (in months): 6
  • Desired start date: 1/03/2022
  • Research unit: Itap, Inrae
  • Contact: Olivier Naud-olivier.naud[AT]inrae.fr
  • #DigitAg: Axis 6, Axis 3, Challenge 3, Challenge 5

The agro-environmental effects that result from spraying plant protection products (PPP) is related to (a) the PPP quantities intercepted by the canopy, (b) the 3D covering of leaves within the canopy, and (c) the losses on the ground or to the air (drift). The sampling methodologies that we have developed to study the quantities that are intercepted in the vine and their 3D repartition involve artificial collectors that are set according to a precise tri-dimensional scheme. Spraying a tracer that is intercepted by these collectors allows for the assessment of deposits in the canopy. The quantitative analysis of deposits data has been performed until now using statistical distributions and without making explicit in the results the spatial information that is attached to the known 3D setting of collectors. Our work has brought forward the importance of the geometrical structure of the canopy and its density as measured by a LiDAR to explain the statistical distribution of deposits obtained on real vine. Research that extend the results is today necessary to lead to a precise and sound agronomical decision that meets new stakes and requirements attached to new PPP (biocontrol notably).
The aim of the internship is to analyse the links between statistical distribution and spatial repartition of deposits in order to build a robust and combined modelling of these two phenomena. This modelling will be very innovative with regards to the state of the art and will involve geostatistics applied in 3D, in the plane of the field as well as in the depth-height plane of the canopy.
It will be valorised after the internship on digitally controlled sprayers in order to build complex and innovative PPP evaluation plans and to optimize the spraying of these PPP.
The newly acquired capacities will foster research collaboration with pathologists and epidemiologists in crop protection within the framework of a precision agriculture that should be ecological as well as digital.

Development of a rape winter stem weevil flights predictive tool, based on the « Vigiculture » database

Keywords : datascience, machine learning, interpretability, digital agriculture, DST, rapeseed, rape winter seed weevil

  • Duration (in months): 6
  • Desired start date: 1/02/2022
  • Research unit: Terres Inovia, Acta
  • Contact: Quentin Legros – q.legros[AT]terresinovia.fr
  • #DigitAg: Axis 5, Axis 6, Challenge 3, Challenge 5

Ceuthorynchus picitarsis larvae can destroy the terminal bud of rapeseed. The management relies on the use of insecticides which target the adults before the beginning of egg laying. This insect can only be observed in the field thanks to the use of yellow traps which attract them to a certain extent. The optimal date of treatments is based on the precise detection of the arrival of the insects in the fields. The ability to predict flights is therefore essential to optimise the use of insecticides in terms of positioning and avoidance of useless treatments. The « Vigiculture » database agregate field observations since 2008. It is used to edit the « Bulletin de Santé du Végétal » (a weekly report about crop sanitary state) that helps farmers in their decisions. Nonetheless, those data are scarcely used to build predictive tools. The database contains more than 52000 observations over 10000 fields between 2008 and 2020 regarding C. picitarsis.
The goal of this project is to build a predictive model based on those data. Several metrics will be considered (date of first flight, daily flight probability, cumulative sum of trapped insects). Observations in the database are geolocalized. It it thus possible to cross the dataset with external sources of information (particularly meteorological ones) that could explains part of the observed variation. The performances of several machine learning algorithms will be compared. The analysis of the best models will provide informations about interpretability that can be compared to field expert observations. New methods currently developed by Olivier GAURIAU for his DigitAg funded PhD thesis will be tested. Their aim is to find the best compromise between prediction performances and interpretability.
As a perspective of this internship, the best model, if it is precise enough will be freely available to users on the Terres Inovia website and could be integrated in future decision support tool

Environnement and life sciences

UAV as an intermediate step for the mapping of vegetation at National level

Keywords : UAV, biomass, mapping , pastoral livestock

  • Duration (in months): 6
  • Desired start date: 1/02/2022
  • Research unit: Selmet, Institut Agro
  • Contact: JB Menassol-jean-baptiste.menassol[AT]supagro.fr
  • #DigitAg: Axis 5, Axis 3, Challenge 4, Challenge 8

Savannas are ecosystems with a strong spatial heterogeneity for both herbaceous and woody vegetation. The quantification of this vegetation (biomass) is a key point in these arid zones for pastoral livestock for example. Field measurements allow to measure biomass over small areas. But how do you get this information over a larger area? Indeed, the pastoral with the high mobility of the animal used the vegetation on large area. Studies show the possibility to establish relationships between field data and open access of medium-resolution satellite images covering the whole country.
In order to perform calibrations between field data and these open access medium-resolution satellite images, it is necessary to perform intense measurements in the field to take this heterogeneity into account.
One possibility is to use very high spatial resolution (VHRS) images as an intermediate step. The calibration of the field data with these VHRS images would allow to:

– produce vegetation maps considering the heterogeneity of the vegetation

– then make a link between these maps and images with lower spatial resolutions.
The UAV is a tool for obtaining THRS images. For 3 years, work has been carried out in Senegal showing the possibility of calibrating drone outputs with measurements of herbaceous and woody savannas.
The objective of this internship is to use this UAV data as an intermediate step between the terrain and satellite images, then to compare the maps produced with current work based on laborious terrain measurement protocols.
This internship will be supervised by Simon Taugourdeau (CIRAD UMR SELMET) and Audrey Jolivot (CIRAD UMR TETIS).

High-throughput phenotyping of functional traits in grapevine to characterize the genetic variability of its responses to abiotic stresses

filled internship 

Keywords : Phenotyping, high-throughput, near infrared spectrometry, wineyard, drought, photosynthesis, transpiration

  • Duration (in months): 6
  • Desired start date: 28/02/2022
  • Research unit: LEPSE, Inrae
  • Contact: Aude Coupel-Ledru – aude.coupel-ledru[AT]inrae.fr
  • #DigitAg: Axis 5, Axis 2, Challenge 2, Challenge 1

Climate change is likely to bring vineyards to levels of water stress that are critical for the production and quality of wines. New varieties are therefore being sought in order to save water while maintaining their photosynthesis, which is required for yield maintenance. Phenotyping these traits in the vineyard on large numbers of grapevines is thus a major prerequisite in order to i) integrate them into breeding programs, and ii) make decision criteria for these traits available to winegrowers. However, the perennial nature of the vine, the cost and the low throughput of conventional measurement methods (photosynthesis/transpiration) clearly limit their deployment at high throughput in the vineyard. The project aims to develop and test the application of new methods of high-throughput phenotyping of functional traits on large populations of interest in order to assess the genetic variability for these traits. Particular interest will be given to the use of near infrared spectrometry (NIRS) and chlorophyll fluorescence as proxies of photosynthetic and hydraulic functions. The work will be undertaken on 2 populations: a half-diallel mating design, and a diversity panel of 279 varieties. The work is divided into two sub-objectives: (a) a calibration phase on a subset of genotypes, on which fine physiological measurements will be coupled to rapid measurements, to establish prediction models of physiological traits (photosynthesis, leaf metabolites, stomatal conductance, etc.) from high-throughput measurements. (b) the deployment of high throughput measurements on whole populations, the prediction of traits of interest by the models established in (a), and the analysis of the variability and genetic determinants of these traits by association genetics.

Assessment of the potential of the Pl@ntNet platform for the identification of pasture species in support of pastoralist management strategies

Keywords : pl@ntnet, Pastures, botanical identification, Artificial intelligence

  • Duration (in months): 6
  • Desired start date: 1/02/2022
  • Research unit: Selmet, Inrae
  • Contact: Pierre Bonnet – pierre.bonnet[AT]cirad.fr
  • #DigitAg: Axis 5, Axis 2, Challenge 4, Challenge 1

The participatory science platform Pl@ntNet offers various web services to help identify plant species from the automated visual analysis of plant photos. The validated data it generates is used for training automated visual classification models, allowing species identification from photos of leaves, flowers, fruits or stems.
Although pastures represent the agricultural ecosystems with the greatest plant diversity, no Pl@ntNet assessment has been conducted on these agroecosystems, which constitute an interesting study model. The species most represented in the pastures, in particular those of the Poaceae family, are not very numerous in the Pl@ntNet learning base. For illustration, it should be noted that less than 100,000 occurrences of Poaceae are visible on the 10 million Pl@nNet observations published on the GBIF site. In addition, pasture plants are most often seen in the vegetative stage making identification all the more difficult.
The first objective of this internship will be to assess the relevance of Pl@ntNet in its current form, for the identification of pasture species. The constitution of a test data set covering around ten species. In a second step, the enrichment of the Pl@ntNet learning base will be carried out with additional data to evaluate the typology and the volume of relevant images for a significant improvement of performances in the studied context, and to allow a level of precision in line with the expectations of the breeders., an evaluation will be carried out in a third step in order to measure the potential of a quadrat approach. Photos of quadrats at a distance from the ground and fixed focal length will be produced, cropped and submitted to the Pl@ntNet identification service.
The intership is supervise by researcher from SELMET and Amap

Human and Economic sciences

Uses of WhatsApp groups for the exchange of experiences among farmers. Case study in Benin.

Keywords : Social network, digital humanities, smartphones, WhatsApp, agriculture, Benin

  • Duration (in months): 6
  • Desired start date: 3/01/2022
  • Research unit: Innovation, Cirad
  • Contact: Paget Nicolas – nicolas.paget@cirad.fr
  • #DigitAg: Axis 1, Axis 2, Challenge 8

Digital technology in agriculture opens the door to many innovations, including an increased capacity to generate and exchange knowledge among peers. Observations made in West Africa lead to two observations. (1) Many of the tools developed are not used (no clear business model, little consideration of the needs and capacities of users). (2) Despite the spectrum of innovations associated with the imaginary of digital agriculture, social networking applications such as WhatsApp (WA) via cell phone are by far the most used. Farmers easily create groups to interact with several dozen or hundreds of people. These groups are formed around various topics (purchases/sales, crop…) or territories and allow the exchange of texts, audios, photos, videos at low cost. However, WA has many limitations (parallel conversations, data storage, complicated information search, difficulty for long exchanges).
The objective is to extend the field of knowledge of exchange of experiences between farmers by focusing on the case of Benin, and by seeking to identify whether it is possible to rely on popular applications, with zero development cost and easy access (financial and capacity) to participate in the exchange, dissemination of knowledge, or the collection of needs in agricultural systems.
How do social networks participate in the exchange and creation of information and knowledge in agriculture?

– What are the uses of social networks like WA by farmers?

– How does this arena compete with and/or complement other sharing and socialization spaces?

Anchored between management and sociology, the methodology will be mixed: observation and characterization of interactions, virtual or direct exchanges with group members, and quantitative analyses of groups.

Digital tools in the design of agroforestry systems. State of the art of practices and needs

Keywords : codesign workshop, agroforestry

  • Duration (in months): 6
  • Desired start date: 3/01/2022
  • Research unit: Agir, Inrae
  • Contact: Julie Labatut – julie.labatut@inrae.fr
  • #DigitAg: Axis 2, Axis 1, Challenge 1, Challenge 5, Challenge, Challenge 8

The design of agroforestry systems, and in particular the design of agroforestry systems, is a complex issue involving multiple stakeholders with different points of view and both implicit and explicit knowledge and experience. Moreover, the effects of the decisions made will only be perceptible in the long term. The research work of the last decade shows that co-design workshops are a relevant way to reach operational solutions. In these workshops, the use of digital technology remains limited but is a promising approach. The development of generalist tools is hampered by the variety of workshop practices and the limited availability of literature on the subject.
This internship aims at analyzing the current practices of agroforestry system co-design workshops in order to identify how digital tools can improve this design. Semi-directive interviews will be conducted with various actors who have had experience in facilitating and participating in co-design workshops to identify the characteristics of the systems that are the focus of the design, the different steps of co-design workshops, the tools currently used and the characteristics expected from a design support tool. The analysis of the needs and means that would allow the development of digital applications in agroforestry system co-design workshops will allow us to answer key questions that are important in the development of digital tools that contribute to system design:

– What visualizations of ecosystem services are most needed?

– What gaps in current tools/methods are most felt by stakeholders?

– Is it necessary, and if so, how should the spatial and temporal dimension of systems be represented?

Digitalization in livestock farming systems: Towards work organizations for the agroecological transition

Keywords : Livestock farming; agroecology; digitalization; work; working conditions; work organization; advisory

  • Duration (in months): 6 mois plus 2 months of short-term contract for paper publication
  • Desired start date: 15/02/2022
  • Research unit: Pegase, Inrae
  • Contact: Anne-Lise Jacquot – anne-lise.jacquot@agrocampus-ouest.fr
  • #DigitAg: Axis 1, Challenge 1, Challenge 5

The recent movement of digitalization in livestock farming requires to update the understanding of the activity conditions of the users. For instance, digital tools are proposed to livestock farmers for an agroecological transition purpose (by example, software for the grazing management) or for improving their work (milking robot, …).
However, developing agroecological livestock farming systems induce the design of new practices to be integrated into various subsystems of the farm, that needs to be changed and rearranged. An adaptation of the work organization, as well as dedicated time, are needed both for the process of agroecological transition and for picking-up the digital tools. These stakes remain a challenge in livestock farming, known as the job with the highest volume of labor time in France. These issues then, call for research to analyze the effects of the digital tools on the work organization, in order to identify the suitable conditions for their use, with the final aim to facilitate the change towards work organization and conditions conducive to the agroecological transition of the livestock farming systems.
The internship, based on an interdisciplinary approach between sociology and animal sciences, will aim at understanding how the digitalization is mobilized by the livestock farmers to adapt their work organization to the agroecological transition. The survey will focus on a diversity of farms in a local territory, based on interviews with a heterogeneity of livestock farmers (with different degrees of digital uses, and different degrees of agroecological development). These interviews will allow to collect the needed data, to analyze their work organization and the role of the digital uses in their work, as well as to determine the obstacles and levers to achieve suitable working conditions and organization both for using digital tools and for developing agroecological practices.

Engineering sciences

Navigation in implication rules extracted from agroecological knowledge in animal and plant health for decision making

filled internship 

Keywords : Agroecology, plant use, Formal Concept Analysis, Implication rule, Visualization

  • Duration (in months): 6
  • Desired start date: 2/01/2022
  • Research unit: LIRMM, Université de Montpellier
  • Contact: HUCHARD Marianne – marianne.huchard@lirmm.fr
  • #DigitAg: Axis 5, Axis 1 Challenge 1, Challenge 3

For a farmer, implementing agroecological practices on his farm requires having a decision support system (DSS) allowing him to identify them. The development of such a DSS requires a sufficiently extensive knowledge base and a knowledge navigation system (SN) adapted to her/his needs, allowing to consider innovative practices, where appropriate.
The Knomana knowledge base, for instance, brings together more than 46,000 descriptions of usage of pesticides and antibiotics plants in plant, animal and human health (Silvie et al. 2021). The visualization platform RCAviz (https://info-demo.lirmm.fr/rcaviz/) has been developed to navigate in this type of knowledge base. Based on Relational Concept Analysis (RCA), a classification method dedicated to relational data, RCAviz makes it possible to navigate graph-type conceptual structures and easily identify, for example, a local plant capable of protecting a crop against an invasive pest, or partly equivalent plants to solve a given sanitary problem.
In addition to conceptual structures, RCA enables to represent knowledge as implications rules, a formalism which, by approaching natural language, is better suited to users in the rural world (e.g. « F_Meliaceae => no-food » : Meliaceae family plants are not consumed as food). The evaluations conducted within the One Health approach (i.e. Mahrach et al. 2021; Saoud et al. 2021) have demonstrated the viability and usefulness of this solution.
The objective of the internship is to develop a software prototype for visualizing knowledge, expressed in the form of rules of implications produced by the FCA4J library (http://www.lirmm.fr/fca4j). This application will allow the rules to be presented in relation to measures of interest or according to a symbolic formulation given by the user, for example the rules relating to a certain set of conditions. This will definitively allow the user to easily exploit them.

Characterization of fruit tree health with RGB imaging using close-range sensing: application to chlorophylle content and to shot hole in peach

Keywords : Close-range sensing, deep learning, machine learning, RGB, horticulture, biotic stress

  • Duration (in months): 6
  • Desired start date: 01-02-2022
  • Research unit: GAFL, Inrae
  • Contact: Morgane Roth – morgane.roth@inrae.fr
  • #DigitAg: Axis 5, Axis 2, Axis 3, Challenge 0, Challenge 1, Challenge 2

Developping agro-ecological practices in fruit tree production implies reducing pesticide use, however to date only few cultivars baring disease and pests resistances are available on the market. Quantifying resistances or tolerances among breeding material or genetic resources in the orchard is challenging: because damages are typically assessed visually with ordinal scale and thus lack resolution and preciseness. In addition, only few tools allow measuring integrative traits (tree vigor, photosynthetic activity) in a quick and reliable way. This master thesis will explore whether data acquired with close-range sensing can allow for the characterization of tree health components and whether these methods can be more efficient than usual ones. To this aim, one specific trait, shot hole symptoms (caused by Coryneum beijerinckii), and an integrative trait, chlorophyll content, were chosen. These traits will be measured with standard methods (visually and with a chlorophyll-meter respectively) and using close-range imaging in two peach orchards managed under low phytosanitary protection. Image acquisition will be done with a RGB camera hold on a PHENOMAN pole. The student will participate to data acquisition and to pre-processing to build a reference dataset. He/she will then estimate relevant phenotypic variables using these images. For shot hole, deep learning algorithms will be trained on manually annotated pictures. For chlorophyll content, machine learning methods will link chlorophyll-meter measurements with one or several variables obtained from images. After a critical analysis of the results, the methods elaborated in this project could be deployed to a larger audience via acquisition protocols and analytic pipelines.

Automatic segmentation of aerial images of agroforestry systems to characterize their structure

Keywords : Image analysis – Classification – Segmentation – Remote sensing – Neural networks – Deep learning – Agroforestry

  • Duration (in months): 6
  • Desired start date: 1/02/2022
  • Research unit: Amap, Cirad
  • Contact: Frédéric Borne – frederic.borne@cirad.fr
  • #DigitAg: Axis 5, Axis 2, Challenge 1, Challenge 5, Challenge 6, Challenge 8

The ecosystem services provided by agroforestry systems are now widely recognized (biodiversity, carbon storage, etc.). However, the richness of biodiversity is difficult to characterize exhaustively in time and space, as is the capacity to store carbon. These quantities are particularly dependent on the structure of an agroforestry system, i.e. the composition and arrangement of the plant resources that make it up.
We propose here to understand the structure of the systems from the analysis of aerial images. However, traditional classification methods have so far only partially met this need for fine characterization of agroforestry systems. They require a lot of expertise, which is difficult to generalize from one image to another. This is why we propose to evaluate the feasibility of automatic classification by convolutional neural networks with the latest methodological developments in the field.
An automatic classification by Deep Learning of aerial images of agroforestry systems in a dedicated environment will be implemented, then compared with a traditional classification on targeted examples.
These application examples concern experimental fields of agroforestry or agropastoral systems, which are located i) in temperate zone on the Restinclières domain (France) – agroforestry systems cereal / walnut, vine / hackberry, breeding zone (horses), … – (ii) in the tropical zone on agroforestry systems for coffee and cocoa in Central Africa. The learning process will be carried out step by step, based on general data, then by FineTuning on a reduced set of already available sets that will be improved step by step by field samples and precise maps available on these study areas (Transfer Learning).