Ph. D. Project
Proposal for a Deep Learning Approach for monitoring Migratory Bird
2020/11/01 - 2023/10/30
Other supervisor(s):
Pr. OUAHAB Kadri ( , Dr. BENYAHIA Abderrezak (
Issue :
Algeria has more than 250 wetlands, 50 of which are classified at the international level for their
importance and ecological role. Wetlands are home to a very wide variety of animal and plant species and
play a major role in biodiversity. In particular, they are privileged places for tens of thousands of water
birds of different species to winter or make a temporary stopover. The decline of certain avian species can
cause serious problems in the food chain and impact environmental safety and even affect public health.
Monitoring the size of bird populations is a common activity for many ornithologists to identify possible
environmental or anthropogenic changes that negatively affect species.
In order to design a bird monitoring system, several questions need to be answered, the first of which is
the definition of the types of birds to be observed requiring the intervention of ornithological experts
associated with the research project. Based on the requirements defined by these experts, the monitoring
system can be specified with several scientific issues that arise: The capture of images by high quality
cameras, the transfer of multimedia data on an adapted network and its storage and the information
processing phase.
This thesis focuses on the study of bird detection which depends directly on the amount of information in
each image and indirectly on the location and position of the cameras and thus also on the characteristics
of the communication network. Finding an ideal combination is a problem that cannot be solved in
polynomial time by exact algorithms of a sequential or parallel nature. It therefore belongs to the class of
problems known as NP-Complete. Solving strategies using artificial intelligence methods are increasingly
applied to this type of problem. Several works carried out in recent years have demonstrated the
usefulness and efficiency of artificial intelligence to solve this combinatorial optimization problem.

Objectives :
We seek to design a deep learning based monitoring approach where the construction of 3D models using
several 2D images plays a central role, as a process allowing the improvement of bird type identification.
We also seek through this work to set up algorithms for the monitoring of complex and dynamic systems
taking into account the communication resources that rely on IoT (Internet of Things) architectures whose
flow rates and lifetime are limited and therefore need to be optimized. For system validation, we propose
the use of a machine learning library created by the Google Brain team called TensorFlow.

Main axes :
1) Make a state of the art of the field of remote monitoring systems and consider its various applications to
better understand their principles.
2) To study the methods of automatic learning.
3) To propose an approach to bird monitoring based on Deep learning with a systemic view: from the
creation of information to its processing through its transport.
4) Implement and test the system on the Annaba site.
Environmental monitoring, Internet of Things, Deep Learning, Cloud computing
Eco-Technic systems engineering