Keywords: multispectral imaging, proximal sensing, semantic segmentation, deep learning,
precision agriculture.
Project description: This internship follows the research conducted by Jehan-Antoine
Vayssade on a multi-criteria approach for discriminating between crops and weeds. The
work has resulted in the development of various methods, both conventional and based on
deep learning, along with several labeled databases. Currently, the initial stages of the
processing chain involve a specific neural network for soil/plant separation and another for
semantic segmentation of leaves. The final classification step uses a conventional approach
with a decision tree to select the most discriminative properties (shape, color, texture, etc.).
The objective of this internship is to enhance the classification step by employing deep
learning.
Activities: Conduct literature review to identify suitable neural networks for the subject.
Familiarize yourself with the existing codebase (Python). Develop and evaluate a deep
learning-based classification method. Contribute to the writing of a scientific article if
applicable.
Expected Output: Develop and evaluate one or more classification methods considering all
leaf properties. Compare with previously developed conventional imaging methods.
Contacts :
Gawain JONES
ImViA – Institut Agro Dijon
gawain.jones@agrosupdijon.fr
tel. : 03 80 77 27 48
Jean-Noël PAOLI
ImViA – Institut Agro Dijon
jean-noel.paoli@agrosupdijon.fr
tel. : 03 80 77 28 18