Observatoire de Paris Institut national de recherche scientifique français Univerité Pierre et Marie Curie Université Paris Diderot - Paris 7

Convolutional Neural Networks for object detection in astronomical images

lundi 12 avril 2021, par David CORNU (LERMA - Observatoire de Paris)

Jeudi 15 avril 2021 à 16h00 , Lieu : En visioconférence

This presentation will cover the basics of Convolutional Neural Networks (CNN) construction and usage. Through simple and illustrative examples, we will discuss why CNN are better suited for images than classical fully connected neural networks. We will explain how CNN work on simple image classification and describe some widely used CNN architectures for object detection inside images.

We will present some of the results of the Minerva team working on SKA Science Data Challenges that consist in detecting galaxies in very large simulated 2D images or 3D spectral cubes. In this context, we will discuss why astronomical images are challenging for CNN object detection methods and present our approach to overcome some of the encountered difficulties.

Pour assister au séminaire, contacter :
To attend the seminar, contact :