DSTN > Projects > Deep4 EnvMonitoring

Researchers in charge :
○ Prof. Jean-Marie Dembele, Professeur Assimilé at Université Gaston Berger de Saint-Louis, ACE MITIC, Senegal.

○ Prof. Eugène Ezin, Professor at the University of Abomey-Calavi, ACE SMIA, Benin.

Other partners:

● Prof. Christophe Cambier, UMI UMMISCO, IRD
● Théophile Bayet, IRD Dakar

DSTN > Projects > Deep4 EnvMonitoring

Title: Deep learning for environmental monitoring; illegal landfills

Type of project: Research project (4 years)

This project addresses environmental management using DeepLearning and Drone imagery; more specifically, the monitoring of small-scale illegal landfills. Indeed, despite the collection of domestic waste by municipal services and its deposit in official landfills, illegal dumps are still found in all African countries. This poses serious health and environmental problems. There is therefore an urgent need to identify these illegal dumps in real time, in order to prevent their emergence and proliferation. Such monitoring, geolocation and characterization can also provide a better way of dealing with illegal dumps at minimal cost. 

This is why we use a convolutional neural network (CNN) to extract features from images or videos and enable recognition. It is well known that CNNs can recognize patterns in a specific frame, but in this project we want to demonstrate the possibility of indirect encoding of weights in deep neural networks composed of thousands of connections. This will enable deep networks for image recognition (such as the one that already detects garbage) to evolve into a hypernetwork bringing together different patterns and being scalable, with the possibility of integrating unknown patterns such as precarious or unfinished constructions, anarchic or illegal land use, crop growth, etc.

The development of automatic flight plans for the drone according to the surveillance objective (small clandestine dumps, precarious or unfinished constructions, anarchic or illegal land use, crop growth...) is another objective of this project, which so far involves partners from Senegal/Benin/France and doctoral/master's students.

● Digital science research topics: Artificial intelligence, DeepLearning, image recognition, NeuroEvolution, evolutionary algorithms.
● Other research themes and application areas: Environment, illegal landfills and land use, satellite images and drones, surveillance and decision-making.