Master2 Internship Offers 2018-2019

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


For 2018-2019, the 17 selected offers deal with anthropology, sociology, agronomy, ICT, innovation, informatics, mathematics, statistics, data sciences, econometrics, electronics, remote sensing…

To apply please send your curriculum and application letter to the relevant contact persons listed below

Locations: Montpellier, Toulouse or Rennes. Some internships are located in Africa and South America.
Allowance: #DigitAg gives internships grants to the research units:  the amount of your allowance is to be ask to contact persons.


Informatics – Semantic Web

Taking into account uncertainties in a decision support tool for the AOP cheese farmer and producer

Keywords: Uncertainty management, decision support system, imperfect knowledge management

  • Internship duration /Starting  : 6 months /March 2019
  • Research Unit:  IATE, Inra Montpellier
  • Contact: patrice.buche [AT]
  • #DigitAg: Axes 4 & 5, Challenges 5 & 7

Cheese making chains valorizing their terroir represent an important economical and agricultural activity in France, with around 17900 milk producers, 1290 farm producers and 432 transformation companies. Cheese making chains enjoying an “Indication Géographique” (AOP/IGP) base their product differentiation strategy on the valorization of local resources in connection with their terroir and on the expression of the know-how of experience as well at the level of the production as the transformation. Internal evolutions to appellations, especially in terms of turn-over and training of operators, strongly weaken the preservation and transmission of these know-how. The development of artificial intelligence methods allowing the exploitation of knowledge bases opens new perspectives in terms of preservation and data management of operational experience, by proposing complex modes of reasoning going well beyond the description and formalization of standard processes. As part of the CASDAR Docamex project (2017-2020), the knowledge engineering team (ICO team) of UMR IATE is developing a method and a tool for decision support (DSS) for the farmer who helps to control a defect or quality of manufacture by recommending the most relevant technological actions to undertake. The DSS also allows, for a given action, to determine all the defects and qualities impacted. These recommendations are based on the formal representation of the causal relationships linking default / quality to actions by explanatory mechanisms. In Docamex, the ICO team works with several AOP chains of different characteristics (Comté, Reblochon, Emmental of Savoy, Salers, Cantal) in order to develop a generic and adaptable DSS. The knowledge manipulated by the ADO is formalized with the Semantic Web languages well adapted to integrate knowledge from heterogeneous sources in the chains. In the master’s project, the methodological lock aimed at modeling the uncertainties associated with the causal relationships linking the defects / qualities to the actions. The first objective is to be able to propose in the ADO a prioritization of actions taking into account these uncertainties. The second objective is to be able to update the uncertainties by taking into account the result of the new cheese experiments recorded in the field. The proposed modeling will be tested by the creation of a software prototype that will be integrated into the DSS.

Agronomy (innovation, economic models, social networks analysis)

Digital uses in short food supply chains: trends, relational dynamics and prospects for the sustainability of agriculture

Keywords: short food chains; digital; uses; social networks; sustainability; participatory research; pluridisciplinarity

  • Internship duration /Starting  : 6 months /April 2019
  • Research Unit: Innovation, Inra Montpellier
  • Contact: gregori.akermann [AT]
  • #DigitAg: Axis 1, Challenges 6 & 7

Short food supply chains are not necessarily new in France but have been renewed since the mad cow crisis and above all their introduction on the political agenda in 2009. Still often associated with activist or protesting forms, they now represent a large diversity of chains, which the use of digital would contribute to develop and to make known, by producers as well as consumers. Players in short food chains, at the same time, are diversifying, with the massive entry of producers from long chains, social entrepreneurs and younger and less-educated consumers: does digital play a role in this direction? Do the equipped chains, mediated by the NICTs, meet the expectations of these new consumers, in terms of information, practicality, guarantees, transparency? If yes, under what conditions? What opportunities and constraints does this generate for producers and entrepreneurs? With which results? This Master, embedded in participatory and pluridisciplinary research, has two objectives: first, from a first inventory made in 2018 by our team in collaboration with Open Food France, an association specialized in the development of digital projects in short food chains, the objective is to better understand the different uses of digital in short food chains and their roles in the change of scale and in the democratization of these chains, based on framing surveys with experts at the national level (among which experts of the RMT Alimentation locale, co-led by INRA UMR Innovation) and in others countries of Europe (within the focus group on short food supply chains of the European Partnership for Innovation to which INRA takes part). Second, through an analysis of different types of short food chains equipped by NICTs, the objective is to study how NICTs redefine the relations and interactions between producers and consumers and with the team leading the chain, which new resources and constraints (information, information of consumers…) they allow to circulate, and how does this contribute to sustainability, of farms especially. In collaboration with Open Food France, and by associating inputs from economic and network sociology and management sciences, we will test two hypotheses in particular: relationships only virtual cannot suffice to build confidence in the chain but make evolve face-to-face relations in a positive direction regarding the economic viability of the structures which carry the chain (farms, enterprises); the relations triggered by the use of digital, moreover, favors the evolution of the short food chain towards taking into account new dimensions of sustainability such as the environment, health or ethics.

Statistics – biostatistics – Modelling

Assimilation of proximal and remote sensing data to improve the forecast capacity of the crop growth model SiriusQuality

Keywords: Decision support, Data assimilation, Crop model, High-throughput phenotyping, Remote sensing

  • Internship duration /Starting  : 6 months /March 2019
  • Research Unit: LEPSE, Inra Montpellier
  • Contact: sibylle.dueri [AT]
  • #DigitAg: Axis 6, Challenges 1 & 2

Crop models simulate the interaction between plants and the environment to estimate yield, harvest quality and the environmental impact of the crop. They represent the daily crop growth as a function of meteorological conditions, water and nitrogen availability and varietal characteristics. Crop models can be used to optimize (strategical and tactical decisions) or to drive (operational decisions) crop management practices, to maximize crop yield and quality and minimize unwanted nitrogen losses (e.g. leaching, N2O emissions). Data assimilation improves the forecast capacity of models and reduces uncertainty of simulations. By collecting spatial information for large areas at low resolution (>10 m), remote sensing provides the means of monitoring agricultural systems at intra-plot to regional scale, to evaluate the effect of different management strategies on yields and the environment. Proximal remote sensing methods (terrestrial or aerial) provide complementary information with higher temporal and spatial resolution. In addition to crop management, proximal remote sensing methods are developed for high throughput phenotyping for varietal selection. Assimilation of fast phenotyping data allows to determine genotype x environment interactions in multi environment experiments, or to analyze the genetic variability of not or punctually measurable characters. The objective of this Master project is to evaluate data assimilation methods to be coupled with a crop model (Sirius Quality; The student will use leaf area index, soil moisture and chlorophyll concentration data from different sources (Sentinel 2, drone, Phenomobile), and will evaluate a hybrid data assimilation method that combines parameter and initial conditions estimation and an original particle filter method (Chen, Trevezas, Cournede, 2014 ; doi.10.1137/1.9781611973273.10) developed in the framework of a collaboration with the Cybeletech agricultural start up. This will contribute to improve the representation of crop growth and reduce uncertainty. The student will be involved in the choice of the parameter and initial condition estimation method, and in the coupling with the data assimilation method. He/She will implement and test the methods for the Sirius Quality model. At the end the developed method will be validated for two additional crop models (STICS and Monica) and will be integrated in the decision support tool developed by Cybeletech.

Applied Mathematics – Agronomy & ICT

Numerical optimization for precision spraying in viticulture: obtaining intervention maps maximizing the protection level associated to reduced doses

Keywords: Optimization, spray efficiency, disease risk, map segmentation

  • Internship duration /Starting  : 4-6 months /April 2019
  • Research Unit: MISTEA, Inra Montpellier
  • Contact: patrice.loisel [AT]
  • #DigitAg: Axis 6, Challenge 3

The wide access to spatialized data at the scale of an agricultural plot as well as technical innovations make it possible to adjust spatially cultural interventions by taking into account spatial heterogeneity. Precision agriculture opens up the possibility of reducing the use of phytosanitary products while preserving agronomic performance, which is a particularly important challenge in viticulture (10 to 24 treatments per year depending on the production context). This Master internship proposal is part of the research action “Precision spraying in viticulture” developed at IFV within the UMT EcoTechViti, and at UMR ITAP and aims to mobilize optimization methods developed at UMR MISTEA dedicated to map zoning (R package ‘Geozoning’). Contrary to the usual scheme of defining action zones by processing vegetation maps independently of cultural actions, we propose here to optimize the very definition of action maps by guiding it with a function representing the level of protection actually achieved and with an estimation of a risk accepted by the vinegrower. This approach would also allow to take into account in-season observations (typically observation of outbreaks of cryptogamic diseases) for the generation of maps. The selected student will implement and test an optimization algorithm defining an action map for precision spraying that i) is based on the modeling of the grapevine protection level and ii) meets the objectives of the vinegrower. The first step will consist in formalizing with crop scientists the problem of optimizing the level of protection from the sprayed dose, the spray efficiency and an the plant density obtained using Lidar data. In a second step, the student will study with mathematicians the use of the optimization method ‘geozoning’ to solve the problem numerically. This step may lead to implement in R or Python a “1d” version of the initial algorithm by considering the curvilinear abscissa of the sprayer path. Finally, scenarios for evaluating the obtained maps will be constructed using data acquired by the UMT EcotechViti. The student will be hosted at UMR MISTEA, supervised by Patrice Loisel and Sébastien Roux and co-supervised by Xavier Delpuech (IFV, UMT EcotechViti) and Olivier Naud (UMR ITAP, UMT EcotechViti). This project will also provide an opportunity for interdisciplinary collaboration on management issues and the economic context at the farm scale. Indeed new questions are raised when using concepts as “phyto budget” (dose to be distributed) and localized risk management. This will be done with colleagues from UMR MOISA (in particular Isabelle Piot-Lepetit and Karine Gauche).

Statistics – Econometrics (mathematical & economical models) – Fluent English is required

Performance and risk in agricultural and agri-food businesses: What can we learn from Business Analytics?

Keywords: Risk, Performance, Business Analytics, Consulting tools

  • Internship duration /Starting  : 6 months /March
  • Research Unit: MOISA, Inra Montpellier
  • Contact: isabelle.piot-lepetit [AT]
  • #DigitAg: Axis 5, Challenges 6 & 7

French agriculture and agri-business firms have been highly decreasing over the last 20 years (concentration and/or disappearance). From 1988 to 2010, more than a half of farms disappeared and, from 1995 to 2015, the number of agricultural cooperatives fell by 40%. This erosion of this economic network of firms as part of the whole value chain did not necessarily lead to better economic and financial performance, considering that the average added value rate of agri-business firms went down in almost 20 years, from 33% (1997) to 18% (2015). It is so particularly worrying. In this context of repetitive economic and financial crisis, markets high volatility, and the increase of firms’ disappearance by closure or liquidation, professionals and public authorities increasingly focused on performance and risks analysis. They show the willingness to prevent firms’ difficulties. The aim of this Master internship’s proposal is to provide an analysis of the relations between performance and risk in agricultural and agri-food companies in France. To do this, the work will consist of the following steps: – A review of the literature in Economics and Management to define the notions of risk and performance as well as characterize the causal relationships already identified ; – Identification of the elements of a Business Analytics approach that can be mobilized to investigate and understand the evolution of the risk and performance of agricultural and agri-food businesses ; – A characterization and classification of existing professional and / or public databases, and ; – A test of the business analytics approach selected. During this Master internship, the concept of Business Analytics will be considered in these three main dimensions: – Descriptive: Use of descriptive statistics and data mining to obtain information characterizing the main trends in performance and risk as well as an identification of variables and relationships that may exist, without ex-ante hypotheses, between the concepts of performance and risk in the collected databases ; – Predictive: Suggestion of models based on longitudinal series to forecast potential future evolutions ; – Prescriptive: Based on optimization models, identification of areas where changes can be expected to improve the performance of agricultural and agribusiness enterprises under risk constraints. During the internship, contacts with both private and public partners in France (APCA, ANIA, Coop de France, Groupama) as well as in the USA (Cornell University and Teradata University Network) are expected. The final objective of the comprehensive project, in which internship topic is considered as a first step, aims to develop a tool as analytical well as consulting services for supporting agricultural and agri-food businesses in preventing of economic and financial risks while maintaining or increasing performance.

Anthropology – Sociology (qualitative surveys)

Do farmers’ representations of cultivated plants influence the way in which digital information and communication technologies are adopted and used in the farm: the case of ICT in viticulture and market gardening

Keywords: ICT, digital information and communication technologies, plants, nonhuman agency, plant water stress sensor, sound diffusion systems

  • Internship duration /Starting  : 6 months /March 2019
  • Research Unit: MIAT, Inra Toulouse et UMR Innovation, Montpellier SupAgro Florac
  • Contact: frederick.garcia [AT], aurelie.javelle [AT]
  • #DigitAg: Axes 2, 2 & 3 – Challenges 1 & 3

The use of digital information and communication technologies (ICT) is growing on farms to help farmers better monitor and control their production systems. These technologies make it possible both to acquire numerous and accurate information on the state of animals or cultivated plants, and to act effectively on their direct environment to control their growth and development. The question of farmers’ adoption of such innovative digital applications is generally addressed in terms of attraction or aversion to new technologies and the organizational transformations they imply for the production system. However, considering recent social science works on the human-non-human relationship, it is also useful to question the links between representations that farmers have of animals and plants that participate in their production system, and adoption and use of digital technologies that impact the development and behavior of these living beings. The work that we wish to develop along this master thesis focuses on the ethnological analysis of the adoption and use of digital information and communication systems in plant productions. We will consider viticulture and market gardening, two areas of application which should allow to observe more “individualized” relations with plants than with, among others, crop systems. We will work on digital tools directly targeting plants, and we will study in particular devices that aim i) to capture information on plants (such as devices for measuring plant water stress in the vineyard); ii) to communicate information to plants (such as sound diffusion systems). This diversity of situations will enable us i) to explore the representations that farmers make of plants for symmetrical communication systems, ii) to understand the diversity of relationships with plants in situations that do not have the same scientific basis: sensor-based technologies relie on relatively stable scientific knowledge, unlike sound devices, which have little established scientific basis. Conducted on about 20 farms, the student’s work will first consist in i) identifying the reasons for the adoption of this equipment and this choice of innovation (sources of information, arguments retained …); ii) analyzing the use of sound devices and the innovations that this brings to the farm (choice of locations, changes in the use of equipment over time, integration into farm practices, etc.). Referring to social science work that explores the agency of plants, the student will then try to understand the representations that farmers have of cultivated plants, in order to identify a possible correlation between these representations and the kind of adoption and usage of digital information and communication technologies by these farmers.

Informatics – Semantic Web

Knowledge representation and reasoning for the design of agroecological cropping systems

Keywords: Agro-ecology – Data and knowledge management – Reasoning – Semantic Web – Artificial Intelligence

  • Internship duration /Starting  : 6 months /February 2019
  • Research Unit: Graphik-LIRMM, Inria
  • Contact: Marie-Laure Mugnier, mugnier [AT]
  • #DigitAg: Axis 4 – Challenge 1

Meeting the agro-ecological challenge leads to design very innovative cropping systems. Achieving this ambition requires both genericity (of the processes) and specificity (of the many possible technical combinations), which implies sharing and integrating various kinds of knowledge to adapt each system to its context (soil, climate, landscape, exploitation). Current agricultural system design methods are based on workshop sessions with experts of multiple domains and skills and rely on the construction of conceptual models (Lamanda et al., 2012) used as an interactive basis to formalize and integrate scientific and empirical knowledge. This master internship fits into an emerging collaboration and is expected to be continued by a PhD thesis, whose final objective will be the construction of a tool (i) dedicated to the elicitation, formalization, integration and sharing of data and knowledge on the functioning and management of agro-systems in the context of the agro-ecological transition of agriculture, and (ii) offering several services that will rely on the semantics of these data and knowledge : exploration and query answering, checking the consistency of the modeling, identifying the elements of a process as well as the relationships between these elements, bringing out the consequences of some changes in the system, etc. This will be a tool for aiding the conception sessions, which will allow the participants to have a global and systemic view of how the studied agro-system operates. This tool will also help to formulate scientific hypotheses, to verify them, and to identify « gaps » in expert knowledge. The aim of the master internship is to carry out the first steps of this project: 1. Start from a small set of conceptual models which describe the agro-ecological functioning of a vineyard plot or an orchard (designed within the ongoing AgroEcoPérennes Casdar project, 2017-2020). These models are about the functioning of perennial plant cultivation systems, centered on the description of processes, specifically concerning diseases and pests. The intern will have to understand, and possibly specify, these conceptual models with the help of agronomists, and study which knowledge representation and reasoning languages (beginning with standard languages from the semantic web) are best suited for their formalization ; 2. Formalize the associated agro-ecological issues and identify generic reasoning problems: exploration / query answering / consistency checking / identification of key elements of a process, computation of the consequences of changes in the system ; 3. Make a proof of concept based on the selected conceptual models.

Informatics – Data Sciences

Development of a method allowing to mine and analyse huge volumes of simulation results from a crop model

Keywords: crop simulation model, multidimensional data warehouse, multicriteria analysis, skyline queries

  • Internship duration /Starting  : 5-6 months /February 2019
  • Research Unit: Lacodam, Inria Rennes
  • Contact: anne-isabelle.graux  [AT]
  • #DigitAg: Axes 4, 5 & 6 – Challenges 1 & 4

With the simultaneous increase in the questions addressed to agriculture, in system knowledge and computing power, simulation models tend to be more and more complex producing increasing volumes of heterogeneous data. These data are generally not fully analysed, due to their huge volume that makes their mining difficult, and to the fact that model users are often just interested in a small part of the data. Futhermore, simulation data are often lost after their valorisation although they could help answering other scientific questions. This suggests a need for a method allowing a storage of simulation data on the long term, as well as an easier mining and analysis of simulation data with the possibility for model users to answer multi-criteria questions. The IRISA-INRIA LACODAM team is developping data mining methods enabling to identify interesting patterns supporting the recommendation of actions. A data warehouse to explore multidimensional simulated data from a agro-hydrological model was recently developed by this team to improve catchment nitrogen management (Bouadi et al., 2017). The objective is to adapt this method for the exploration and analysis of the simulated data from the STICS crop model (Brisson et al., 2003) that were produced in the framework of a French study called « Production, exportation d’azote et risques de lessivage » (Graux et al., 2017). STICS simulates crop production and associated environmental risks and benefits at the plot and crop rotation scales, according to soil and climate conditions and to crop management. The work has already started with the development of a relational database where simulated data can be stored. The first step of the training is to move this database towards a multidimensional data warehouse involving time and spatial dimensions and allowing an easier exploration and interactive analysis of the stored data. Based on this data warehouse, the trainee will be in charge of proposing and developing a method allowing the analysis and retrieval of multidimensional information, based on skyline queries. The latter offer the possibility to use user’s preferences to detect and bring out possibly interesting data (i.e. skyline points) and to identify multi-criteria trade-off solutions. The method has also to allow explaining to users the result from a skyline query (Chester et al., 2015) (i.e. why a situation which is a priori not interesting is included in the query’s result and conversely, why a situation wihch a priori interesting has been excluded from the query’s result). This will enable to put users at the heart of decision-making and to co-develop accepted recommendations.

Statistics – Programming & scientific computation with R, Matlab or Scilab

Quality characterisation of crowdsourcing data : the use case of territorial management of vine water status

Keywords: crowdsourcing, spatial data, data quality, viticulture

  • Internship duration /Starting  : 6 months /April 2019
  • Research Unit: ITAP, Irstea Montpellier
  • Contact: james.taylor  [AT]
  • #DigitAg: Axis 5 – Challenge 6

The development of smartphone ( is creating new opportunities for crowdsourcing applications in agriculture (Kobori et al., 2016 ; Silvertown, 2009).The strength of collaborative data based projects is the vast number of contributors generating a considerable amount of data. This strength is also the main weakness of these methods. Indeed, the quality of gathered data may be impacted by the diversity of contributors having highly variable levels of involvement, skills and observation effort (Kosmala et al., 2016). The quality of crowdsourced data necessarily impacts all the data processing chain and, at the end, the quality of predictions made on the variable of interest. Characterising the quality of crowdsourced data is therefore a major issue and a prerequisite to the development of these methods (Pipino et al., 2002). The proposed master degree internship will focus on a crowdsourcing approach mixed to existing approaches for the territorial management of vine water status. (Baralon et al., 2012, Martínez-Vergara et al. 2014). This use case is relevant for two main reasons: – It is exemplary of crowdsourcing methods applied to agriculture with spatial and temporal phenomenon ; – It is strategic for viticulture in a context of global warming. Indeed, crowdsourcing may be an approach complementary for professionals to existing decision tools. The research team has worked on this issue for several years. It has developed empirical modelling approaches (Baralon et al., 2012, Martínez-Vergara et al. 2014) that can be transferred to crowdsourcing data. Moreover, the team has built a large database on several territories to validate the methodologies developed during the internship. The work will focus on an important limit of crowdsourcing data: the characterization of data quality. The goal of this work will be to explore an original approach based on the specificities of crowdsourcing data in agriculture: – The spatial consistency of data: on a given neighbourhood, data have similar frequencies or values; – The temporal consistency of data : the temporal dynamic of observed phenomenon has a trend determined by the evolution of the season, the climate, etc.; Those specificities may be used to detect suspect data (outliers) and to characterise the data quality. The chosen approach will be based on classic statistical and geostatistical methods applied to spatial and temporal dynamics. This work is interdisciplinary because it is at the interface of agronomy, statistics and geostatistics. It may result on a PhD including a strong collaboration with several research teams (UMR Mistea, Lirmm …) and professional partners (IFV, CA34, companies of chaire AgroTIC).

Informatics – Statistics – Compturer programming with R and Shiny

Integration of statistical analysis and visualization tools for the global agroecology information system AEGIS

Keywords: Data analysis, statistical design ,information system, visual analytics

  • Internship duration /Starting  : 4-6 months /August 2018
  • Research Unit: AIDA, Cirad Montpellier
  • Contact: sandrine.auzoux [AT]
  • #DigitAg: Axes 4 & 5 – Challenge 1

AEGIS (Agro-Ecological Global Information System) is positioned as an information system that supports digital agriculture and the successful transition to agroecology. This tool offers functionalities that are useful not only for researchers and students but also for industries, technical institutes and farmers, whether in Northern or Southern countries. AEGIS’ priorities are: rationalizing the observation function, sharing and standardizing the data collected. The aim is to strengthen and perpetuate observation systems for agro-ecological resources and their uses. AEGIS must be able to produce homogeneous data at different scales of observations by meeting industrial quality standards, to develop its tools and diversify the indicators and strive to meet the expectations of both research and experts but also environmental policies, and this, through: (i) the development of generic statistical analysis tools, (ii) the implementation of ex-ante and ex-post data processing methodologies, (iii) the provision of data sets for simulation and (iv) the development of complex visualization tools to facilitate the interpretation of data and to highlight indicators, patterns, trends and correlations inaccessible from raw data. By aiming at the integration, through dashboards, of different data analysis, processing, simulation and visualization tools, aegis aims to be a complete steering and decision support tool in the context of agroecosystems. The objective of the course is to create a set of statistical treatments allowing a first approach exploration and visualization of the data contained in AEGIS. Initially, a needs study will be conducted in interaction with a panel of agricultural researchers from CIRAD, as well as its public and private partners (INRA, IRD, IRSTEA, SupAgro, SASRI, CGIAR, etc.) whose project data are already in AEGIS, in order to establish a list of statistical treatments to meet most of the needs expressed. In a second step, the associated R scripts will be created. Finally, it will be necessary to integrate these scripts into the AEGIS framework. These last phases will be realized using shiny ( – Shiny is an R package that facilitates the creation of interactive Web applications directly from R). This course will take place in a heterogeneous and interdisciplinary scientific environment. It requires a good knowledge of R language and shiny. Some notions on relational databases and on the MVC architecture widely used for web applications would be appreciated.

Informatics   Mathématics – Programming with C++, Java & Python

dRdN, Raphias and Neurons, a neural network for counting raffia in the wild

Keywords: Classification, neural network, image processing, palm trees counting

  • Internship duration /Starting  : 6 months /March 2019
  • Research Unit: AMAP, Cirad Montpellier
  • Contact: frederic.borne [AT
  • #DigitAg: Axes 2 & 5 – Chalenges 6 & 7

Palms are one of the most useful plant families, providing many economic, ecological and societal benefits in the tropics. The species of the genus Raphia is the most widely used and important in Africa. However, anthropogenic pressure has negative impacts at the local and regional level on Raphias and no estimates of basic parameters (number of individuals, biomass, etc.) have been undertaken. Under natural conditions, these Raphias gather in swampy forests where the crowns are nested and therefore difficult to count in aerial view. To date, no study has been done on their enumeration despite their societal importance. Recent palm counting studies concern only oil palm plantations and use “Deep Learning” approaches. The proposed internship will focus on this “Deep Learning” approach by deploying a neural network specifically trained for the detection and enumeration of Raphias in natural environments on remote sensing images. For this, the trainee will have at his disposal two sets of images: a set of virtual reference images consisting of simulated images from models of Raphias areas modeled using AMAP’s Xplo software and a set of real images (UAV images and THRS satellite images) which will serve as validation test. First, the trainee will have to make a state of the art on the “Deep Learning” methods best adapted to the processing of remote sensing images and to our subject of study. Then, it will deploy and configure the selected network and apply it to both data sets. The evaluation of network performance will be conducted by (i) estimating the loss function (overall error), (ii) conventional statistical indicators (measurement) and (iii) geometric indicators (overall Dice index). The supervision will be carried out by the i2p theme (UMR Amap), the UMR DIADE and by methodological support from UMR Tetis.

Agronomy & ICT – Informatics

Specialisation of a location and segmentation neuronal network for the implementation of a mango yield estimation tool based on image analysis

Keywords: Machine-learning, convolutional neural networks, mango detection, yield estimate, Senegal

  • Internship duration /Starting  : 6 months /March 2019
  • Research Unit: HortSys, Cirad Montpellier
  • Contact: emile.faye [AT]
  • #DigitAg: Axes 2, 3 & 6 – Challenges 5 & 8

The estimation of pre-harvest agricultural production is mandatory to face development challenges and to reduce the farmer vulnerability to global changes. Indeed, one of the main issues that currently hampers mango crop development is the lack of tools for accurately estimate crop yield at relevant stages of growth to inform farmer decisions. To date, yield estimation in tropical orchards is still based on a field visual inspection of a limited sample of trees, which is a tedious and time consuming method that relies on the observer’s reliability and accuracy (Linker et al., 2012). Precision agriculture, and in particular the processing of visual data flows, offers new opportunities to gather accurate and relevant information for early fruit yield estimation. In this context, Convolutional Neural Networks have recently shown their ability to detect fruits (Bargotti et al., 2016, 2017), and particularly for mango trees (Sa et al., 2016). If the CNN “knows” how to count fruit, the research question concerns its actual capacities to qualify the fruits, i.e. to identify among the fruits detected in the images those which will actually be harvested by the farmer. In this context, the proposed internship fits and complements an ongoing #DigitAg doctoral thesis on estimates of mango yields in West Africa. Supported by the “Evaluation and design of sustainable horticultural systems team” team of HortSys Lab, the internship will be conducted in collaboration with the group “Imagery of Plants and Landscapes” of the Amap lab, (and the methodological support of the team “Image and Interaction” of the LIRMM) (Borianne et al., 2018). The trainee will work on a set of images captured in Senegal in 2017-2018 on three specific points with high scientific added value in digital agriculture: – the specialization of the Fast-CNN network (Ren et al., 2017) for the automatic counting of fruits in pre-harvest stages and their phenological qualification; – the evaluation of the efficiency of the predictive capacities of the network, thereby addressing complex (and often taboo) questions of the implementation of the validation and the “ground truth” sets; – the segmentation of detected mango surfaces to assess their level of harvestability. The results of this internship will firstly provide the #DigitAg doctoral thesis to analyse mango yield variability according to the structural parameters of the trees. In a second step, they will lay the foundations of an operational tool for the producer for the counting of mangoes on the trees before the harvest. This mango yield estimation tool based on image analysis could be further transposed to other speculations.

Agronomy & ICT – Innovation – Fluent Portuguese is required (surveys)

The use of smartphones in the Açaí sector in the Brazilian Amazon

Keywords: Açaí, smartphones, prices, quality, supply chains, Brasil

  • Internship duration /Starting  : 6 months /May 2019
  • Research Unit: Innovation, Cirad Montpellier
  • Contact: nathalie.cialdella [AT]
  • #DigitAg: Axis 1, Challenges 7 & 8

The use of multifunctional mobile phones (“smartphones”) is spreading among farmers around the world. The different uses that are made of it (access to information on market prices and technical advice, announcement of availability of products, organization for collective deliveries, discussion forum…) have some effects on the agricultural sector. Studies on this subject are still unusual (Baumüller, 2017). As part of the Açaí’action project (building knowledge and consolidating quality markets for Amazonian socio-biodiversity products), the research training’s objective is to collect information on the use of multifunctional mobile phones (“smartphones”) by actors in the Açaí sector. The study will focus on producers who harvest açaí in order to measure the impact of these new practices on the prices and the quality. Through smartphone applications (especially WhatsApp), the prices on the places of exchange are instantaneously shared and reach even isolated producers. This news way of communicating and informing is likely to improve the producers’ ability to negotiate, so far very limited and dependent on intermediaries (Pegler, 2011). The trainee student will have to design a questionnaire, to carry out the survey within two groups of açaí producers (one group using smartphones and one control group), then to analyze the results on the effects of using smartphones on the price and the quality. She or he will have to be fluent in Portuguese. Field research will be the surroundings of the city of Belém, in Brazil. This city is indeed the main trading center for fruit (90% of national production) and processed products (nearly 50% of national production) (Homma et al., 2006). In Belém, there is a diversity of supply chains: some buyers apply a preferential price to ensure the quality of the fruits and foster the producers (Cialdella et al, 2017).

Informatics – Remote sensing  [automatic learning methods, automatic classification of multi-sensor satellite data]

Transfer Learning for the adaptation of deep learning methods for land cover mapping of Southern agrosystems

Keywords: Remote sensing, agricultural land cover, southern countries, deep learning, transfer learning

  • Internship duration /Starting  : 6 months /April 2019
  • Research Unit: TETIS, Cirad Montpellier
  • Contact: raffaele.gaetano [AT]
  • #DigitAg: Axis 5 – Challenge 6 & 8

The production of spatialized information on farming practices in agricultural areas of the southern hemisphere is currently a major development issue for countries at risk of food security. Initiatives such as GEOGLAM (Group on Earth Observation – GLobal Agricultural Monitoring) are increasingly advocating the use of earth observation data and tools to answer to this challenge. Indeed, the availability and variety of remote sensing data from satellite missions, with growing spatial resolutions and acquisition frequencies, make it possible to accurately monitor agricultural activities, even on larger scales. Nevertheless, the automatic methods of agricultural land use mapping from remote sensing data often require a considerable amount of reference data for their calibration, and in view of the increasing volume of images to process this availability is generally limited in Southern countries, since their acquisition is often difficult and expensive. This is the context of this internship. We propose to leverage some land use mapping methods recently developed at TETIS, based on Deep Learning, and to adapt them to the agricultural contexts of southern countries while exploiting the wide availability of information on cultivated areas in the North. More specifically, our goal is to exploit the possibilities offered by Deep Learning in terms of “transfer of models” between different contexts: we therefore propose to set up a Transfer Learning strategy that will allow pre-training of classifiers using reference agricultural data in France, so as to orient the data analysis towards the specificities of agriculture and, secondly, to refine these models with the help of the available reference data on a study site in the South, which are normally not sufficient for a complete calibration when using Deep Learning methods. We will design and develop this Transfer Learning strategy for agricultural land use by coupling time series of images from the ESA Sentinel-2 mission (10 meters of spatial resolution, 5 days of repetitiveness) with scenes at Very High Spatial Resolution (order of 1 meter), such as SPOT6/7 or Pleiades. The Sentinel-2 time series are freely accessible from the ESA platform, and images at VHSR (Spot 6/7) will be made available by the GEOSUD team. This new method will be tested on a study site in Burkina Faso (commune of Koumbia, in the province of Tuy), on which several land surveys have been carried out in recent years by the UMR TETIS (Cirad), with the support of one or more pre-training zones located in Metropolitan France.

Agronomy, Wastewater treatment (reuse) – Informatics, Applied Mathematics

Operationalization of an optimization model to the re-use of treated waste water for crop fertigation in Mediterranean conditions

Keywords: Treated waste water reuse, nutrients, optimization, sensibility, calibration

  • Internship duration /Starting  : 6 months /March 2019
  • Research Unit: ITAP, Montpellier SupAgro
  • Contact: carole.sinfort [AT]
  • #DigitAg: Axes 2 & 6 – Challenges 1, 6 & 8

Re-use of treated waste water is a strategic opportunity to save water but also for crop nutrients (mainly nitrogen and phosphorus). Implementing this principle reveals optimisation and applicability issues. The optimisation concerns 2 points. The first one is to adapt waste water treatment to fit with plant needs in a dynamic way. The second one is to drive the outflows to fulfill plant needs while avoiding nutrients storage in the soil (and leaching). To reach this objective, a model platform is currently developed in a #Digitag post-doc project (opti-reuse). This platform is based on the principle of double modelling: basic processes models (bio-reactor, transport, crop-soil climate system) are simplified in their domain of use and feed an optimisation model to drive stages of the treatment process and inputs to the crops. The objective of this master project is to bring answers to the applicability and the impact evaluation issues building on two case studies: the pilot site of Murviel-lesMontpellier (34), identified in the opti-reuse project and the pilot site of Saint Martin de Castillon (84). Both sites are equiped for the reuse or treated waste water for crop fertigation: olive and vines in Murviel les Montpellier, durum wheat and melons in Saint Martin de Castillon. In the first site water is treated through lagooning and tertiary treatment with a membrane bio-reactor. In the second site, treatment is made with a bacterian bed supplemented with a sand filter and UV beam for the tertiary treatment. Then both sites are complementary considering the treatment process and also the irrigated cropping systems. This master projects will be structured around the following steps: 1. To participate to the identification of scenarios for the calibration of the models in both sites (one per site) ; 2. For these scenarios, to proceed with a sensibility analysis to identify predominant parameters ; 3. To identify measurements to be made for the calibration: element to be measured, sampling, calendar, experimental designs ; 4. To identify and characterize the errors of the simplified models and to propose improvements ; 5. To provide accurate inventory data for the impact evaluation of the two scenarios ; 6. To define a methodology for the calibration and the implementation of the platform for future scenarios or case studies.

Electronics – Microelectronics

Development of an implantable electronic device for the monitoring of fish gonad development

Keywords: Aquaculture, pêche, reproduction, gonade, dispositif électronique implantable

  • Internship duration /Starting: 6 months /February 2019
  • Research Unit:  LIRMM, Université de Montpellier
  • Contact: Vincent Kerzerho, kerzerho [AT]
  • #DigitAg: Axis 3 – Challenge 4

This internship aims at initiating the development of an electronic device and its associated electrode for the implantable monitoring of the development of fish gonad. The challenge is to develop a new measurement technic for a physiological parameter monitoring. Indeed, current approaches consist in repetitive manipulation of fishes for a visual analysis of explanted oocytes. The purpose of such device is to 1) enhance selection programs in the context of aquaculture 2) enhance knowledge of wild species for better stock management. The development of such innovative device will follow three consecutive phases, 1/ in-vitro measurement, 2/ in-vivo out-of-water measurement 3/ in-vivo in-situ measurement. The in-vitro measurement phase aims at defining the suitable electrical measurement method and the associated signature of the different development steps of gonads. The calibration of the electrical measurement will be based on oocytes extracted from seabasses with different stages of gonad development. The gonad development stages will be evaluated by visual measurement of the size of oocytes and by hormone dosing (e.g. vitellogenin). There is a challenge in developing a 2D electrode for in-vitro measurement which enable to identify the best electrical signature of the targeted physiological parameter. Based on these well-controlled measurements, a first electrical model of the interface will be defined to support the next development phase. The electrode will be fabricated at the clean room of the university of Montpellier. The next development phase will be based on in-vivo out-of-water measurements. It aims at developing an electrode taking into account 3D characteristics of the gonad while guarantying electrical and mechanical properties of the device. The topology of the electrode and the characteristics of the electrical pads will need to minimize the influence of the variability of oocyte developments inside one gonad. Fishes will be not sacrificed. Then it will be requested to develop a technique for temporary implantation of the electrode. Thanks to these measurements, the electrical model of the interface will be enhanced considering the 3D environment of the measurement. The last development phase will be focusing on the implantation of the electrode for in-vivo in-situ measurements. In this context, the impact of the device on the behavior of the fish will have to be limited. The electrical circuit needed for in-situ measurement will have to be developed. Strong constraints remain to limit its impact on the fish. As a consequence, the electronics will need to be integrated. The device will be validated on 3 species: sea bass, sea bream, platax.

Biostatistics – Data analysis, modeling, programming (R / Python / Matlab…)

Data analysis of real-time body temperature measurements in pig: data mining approach to predict animal responses to heat stress

  • Internship duration /Starting: 6 months / March 2019
  • Research Unit: Inra, UMR PEGASE (Rennes)
  • Contact : david.renaudeau [AT]
  • #DigitAg : Axes 2, 3 & 5 – Challenges 2 & 4

Affordable measurement of core body temperature in a continuous, real-time fashion is now possible in livestock. Body temperature is a key physiological parameter that provides important insights into the study of thermoregulation, physiology and behaviour or responses to environmental change Up to now, the reference method for monitoring internal temperature was based on punctual rectal temperature measurements by using a medical thermometer. This method requires the immobilization of the animal with possible bias in the body temperature determination due to the animal stress. We recently validated in pig a new system for a continuous monitoring of internal temperature based on the use of telemetry pills. This system is composed of an implantable pill that wirelessly and continuously transmits BT to a dedicated recorder. Our works showed that changes in internal body temperature in pigs can be explained by factors related to livestock conditions (especially climatic conditions) and / or by factors related to the animal (level of physical activity, feeding behavior, growth potential, etc.). This continuous monitoring of internal temperature associated with an ad-hoc mathematical analysis could be an interesting source of information in the future to assist the farmer in managing his herd. The objectives of this studentship are 1/ to initially identify different profiles of nycthemeral body temperature variation, 2/ to investigate the association between profiles of temperature changes and heat tolerance and 3/ to improve prediction of heat tolerance from body temperature profiles previously measured in thermoneutral conditions. To carry out this work, the student will have an access to a database with individual data of body temperature (1 measurement/min) obtained on a total of 60 animals submitted to experimental thermal challenges during the growing-finishing period.

Voir aussi : les offres de stage proposées par les entreprises membres de #DigitAg (à venir)