Paola, Jun, Boris and Shiyu, international master’s degree trainees

In the summer of 2018, the first #DigitAg Master 2 trainees will have completed their internship. On the occasion of the #DigitAgora (#DigitAg Lab gathering), several of them presented their subjects and first results. International students hosted in #DigitAg Lab member laboratories testify below to their stay in France.



Paola Benedetti

In Italy, Paola is a master student in computer science at UNITO (Università degli Studi di Torino). She arrived in France in 2017 as part of the Erasmus+ program. After studying at the University of Montpellier, Paola applied for a master2 internship, funded by #DigitAg. She is supervised by Dino Ienco, who collaborates with Ruggero Pensa, hier supervisor in Turin.


I graduated from the European School, and then I went on to studying computer science. I’m finishing my studies in virtual reality and multimedia at the University of Turin. I had wanted to make my Erasmus application for years, and I obtained a grant from my university last year.


Aware of my areas of interests, Professor Pensa, coordinator of Erasmus exchanges with France, put me in contact with Dino Ienco. At the University of Montpellier, I noticed a significant difference in the approach to group work, internal and external interactions on work are encouraged. I could not have found better!


These 6 months in France were intense. I met people who shared their passion for their work which help me improve and do better. On a personal level, I was “thrown” into an international environment. I met many people from different backgrounds with whom I shared experiences, interests, funny misunderstandings due to language issues… It was a unique experience, a period that ended but changed my perspectives on the future. It changed me and continues to do so!


Why did I apply? Working for 6 months on a complex subject such as neural networks, learning about methods on satellite images processing, and this, in a stimulating research environment, was an opportunity not to be missed! During my studies, I did not have the opportunity to work on satellite imagery in agriculture. I wasn’t familiar with digital agriculture and so, I had no idea that it was such a vast area of ​​research, I was hooked…

What stage are you at?  We have achieved our goals, with the development of M3Fusion, a deep learning model which effectively exploits heterogenious informations from different sensors by merging remote sensing data. These results are currently beeing valorized through a presentation at a conference and a submitted article  (an extended version of the article was submitted to IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing).

After the master? I am looking for a job in a research environment to continue in this direction and then apply for a PhD. I would like to work in the field of computer vision by exploiting machine techniques and deep learning.

  • Subject: Multi-Scale and Multi-Temporal satellite data fusion for land cover mapping via Deep Learning
  • Abstract: Nowadays, the quantity of available data produced by satellite mission is increasing exponentially. The possibility of acquiring several satellite images on the same zones during the time allows to generate time series of satellite images to make possible to monitor agricultural crops dynamics over one or several years. For example, the Sentinel space mission, through the Sentinel 2 satellites, can produce very high temporal resolution satellite image time series (every 5 days) with a spatial resolution of 10 meters. How to effectively manage and analyze these huge quantities of data in order to improve the characterization of agricultural practices is still an open challenge. We propose a master internship to design and develop automatic innovative methods to predict land use (in an agricultural context) from the integration of time series of satellite images with high temporal resolution (i.e Sentinel 2 – at 10 meters) with satellite images (i.e SPOT6/7 and / or Pleiades) with very high spatial resolution – VHSR (between 0.5 and 1.5 meters) but with a low temporal resolution (one or two images per year). The Sentinel 2 time series as well as the VHSR images (Pléiades and Spot 6/7) images will be available through the GEOSUD and THEIA project. To deal with the huge amount of available data, we will tackle the data integration issue, considering different spatial and temporal resolution for land-use prediction, with Deep Learning techniques such as Convolutional Neural Networks (CNNs) and Recurrent Neuron Networks (RNNs). The RNNs will be used to model and analyze the optical remote sensing time series given their ability to model temporally correlated signals while; THRS images will be analyzed using CNNs methods which are more adapted to deal with spatially correlated data. The integration between the two data sources, with the aim of predicting agricultural land use, will be achieved by combining the outputs of the two types of neural networks (RNNs and CNNs).
  • Communication: P. Benedetti, Dino Ienco, Raffaele Gaetano, Kenji Ose, Ruggero G. Pensa, Stéphane Dupuy (2018). M3Fusion : Un modèle d’apprentissage profond pour la fusion de données satellitaires Multi-Echelles/Modalités/Temporelles. Conférence Française de Photogrammétrie et de Télédétection (CFPT), Marne-La-Vallée, 26-27 juin. 2018
  • Supervisor ; Dino Ienco, Irstea TETIS
  • #DigitAg :  Axis 5 – Challenges 3 & 8

Jun Zhan

Jun has a Master 1 in Statistics from Shandong University. She arrived in France as part of an exchange project between his university and the University of South Brittany. She then enrolled in a Master 2 in Mathematics – Statistical and Data Science at the University of Grenoble-Alpes. After some initial courses that were a bit difficult to follow, Jun appreciated the diversity of the courses offered in France: “Many of the basic courses are similar, but the courses are more detailed, with choices of interesting subjects and more practical courses and projects.”

Jun found the subject on the website of the French Society of Statistics.

Why did you apply? “It’s a big data project and I would like to work in this area, I wansn’t familiar with digital agriculture before starting, it’s great, it can help us solve many practical problems.”
Benefits of this internship? “I know more about digital agriculture and I have learnt new methods of data analysis, I have gained experience now.”

  • Subject: Detection of state from massive and temporal data of movements of cows
  • Abstract: The management, operation and analysis of large datasets produces new values ​​in a growing number of fields (financial, commercial, scientific, etc.) leading to major transformations in the practices concerned. Agronomy, by means of increasingly reliable observation devices (sensors, imaging) is also changing in this direction.This course aims at the extraction of new forms of knowledge concerning cows in buildings, thanks to the data of geo-localization that are recorded daily and at high frequency (1.6 Hz). In order to detect changes in the state (disease, heat, etc.) of bovines, we propose to work on two interdisciplinary approaches, using mathematical modeling, inferential statistics and computing. The first outcome of the work on stochastic modeling carried out within MISTEA proposes to model the different possible states by a centered individual model focused on the detection. The second one relies on the matrix profile and allows to identify repetitions of patterns between time series or within a series. Faced with mass data problems (160 dairy cows x 144,000 positions (x, y) x 180 days), these two complementary approaches both require data distribution methods and parallelization of algorithms and models. The objective is to adapt these two approaches to cow data in order to anticipate signs of interest on the condition of bovine animals. Finally, in a more exploratory part of this work, it will be necessary to envisage the possibilities of combining the two approaches. A first track would be to study how. On the basis of the information provided by the modeling, we can target and improve the data analytics (matrix profile) approach.
  • Co-supervisors; Bertrand Cloez (Inra MISTEA), Reza Akbarinia & Florent Masseglia (Inria Zenith)
  • #DigitAg:  Axes 5 & 6 – Challenge 4


Boris Biao Babatoundé


Boris Biao is a Benisois student of the ECODEVA (Environmental & agricultural development and food economics) Master at Montpellier Supagro, Boris is an agronomist, graduate of the Faculty of Agronomy of the University of Parakou in Benin:  “I camed to France to pursue my studies and to improve my knowledge on agricultural transition support tools as well as on scientific research. I applied immediately because I’m passionate about digital agriculture innovations. I have some basic ICT skills applied to agriculture, and now I can better see their huge range of potential applications.”


About his studies at Montpellier SupAgro, Boris emphasizes that “the most important thing is to discover other ways of learning and interacting …. here the teaching is perfect.”

  • Subject: Description of the institutional context of the sectoral innovation system of digital agriculture
  • Abstract: Today, digital technologies are emerging as a means of improving agriculture in terms of productivity, environmental protection, working conditions of farmers and traceability of food to the consumer. As such, they are one of the priority objectives for innovation in French and European agricultural policies. Thus, digital agriculture is a strategic issue for France as described in the report Agriculture innovation 2025 (Bournigal, Houllier, Lecouvey, Pringuet, 2015) as well as in numerous reports published by European bodies (European Parliament, 2016, for example). The objective of this work is to provide an overview of the institutional environment of digital agriculture at the national and European levels, based mainly on neo-institutional theories. The aim is to highlight: – The institutional environment of digital agriculture; – The institutional framework of digital agriculture; -The process of institutionalizing digital technology in agriculture. In order to understand the various institutional pressures, isomorphisms (normative, coercive or mimetic) at work (DiMaggio and Powell, 1983), several questions can be asked:
    – Which institutions are involved (Friedberg, 1998)? – Which policies, packages and measures are aimed at promoting the development of digital innovations in agriculture? – Who are the stakeholders involved, from producers of technologies, research-related, to farmers and other users in food systems, including prescribers, and also agricultural advisors? The theoretical framework may be based on the notion of “sectoral innovation system” defined by Malerba (2002) as ” a set of products and the set of agents carrying out market and non-market interactions for the creation, production and sale of those products ». This framework should make it possible to identify categories of tools, according to their characteristics, their specific uses.
  • Co-supervisors; Leïla Temri (Montpellier SupAgro MOISA) et Nina Lachia (Montpellier SupAgro AgroTIC)
  • #DigitAg: Axes 1 & 2 – Challenges 5 & 7


Shiyu Liu

In China, Shiyu graduated from Northwestern Polytechnic University in Xi’An (Shaanxi Province) with a degree in mechanical engineering. He arrived in France in 2015 to pursue engineering studies thanks to the dual degree program between his university and INSA Lyon. He then continued with the Master 2 Robotics from the University of Montpellier. Like Jun, Shiyu has noticed that in France, studies, especially engineering studies, insist on putting knowledge into practice through internships or research projects: “In China, we have fewer collaborations with companies or laboratories when doing studies.” He appreciated the multidisciplinary nature of the courses and the internships. What can be improved? “The admission procedure, the planning of courses and projects, and the validation of school years”.

Shiyu is particularly interested in this subject because he has taken a lot of courses and carried out several research projects in the field of robotics and computer vision. He was made aware of the internship offer by the teachers of his master: “I had heard about digital agriculture but I had no knowledge on the subject. Now, I think the techniques must be developped and implemented in agriculture applications. This is an important theme for our future.”

Where are you in your research?  “I have developped algorithms and I am in the process of optimizing them for a shorter image processing time.”

What’s the next for Shiyu ? “I’m looking for a doctoral thesis topic, in the same field of robotics or computer vision.”


  • Subject: Spectral band registration by 3D reconstruction for short range operation of agricultural multispectral imaging sensors
  • Abstract: Application of UAV to crop monitoring, which is rapidly increasing, relies on the availability of multispectral imaging sensors combining visible and near infrared spectral bands. Nowadays, various commercial devices are proposed (4-bands camera Sequoia (, 6-bands camera Airphen (, which are all based on the assembly of elementary cameras equipped with their own lens. This multi-lenses solution is satisfactory for the acquisition of images with a spatial resolution limited to a few centimeters per pixel. However, it is not usable for applications requiring a better resolution, such as weed detection, emerging plant counting, spot disease detection or very close monitoring. Indeed, at short range, the uniform superposition of images corresponding to the different spectral bands is impossible due to parallax effects. This subject addresses the development of a registration method between spectral bands adapted to short range imagery, relying on the 3D reconstruction of the targeted scene.
  • Supervisors; Gilles Rabatel, Irstea ITAP
  • #DigitAg: Axis 3 – Challenges 1, 2 & 3


See also: