Innovative techniques to find strongly lensed systems

Abstract

A galaxy-galaxy lens is a phenomenon in which the light of a background distant galaxy is deflected in the vicinity of a massive foreground galaxy. The occurrence of this phenomenon is very rare. The advent of big-data surveys is an opportunity to detect gravitational lenses only if the proper tools are built. This work aims to build such a tool by testing a set of innovative techniques using parametric and non-parametric models to identify the presence of lensed galaxies in a dataset. The dataset used in this work is a simulated dataset combining true galaxy images with lens simulations also based on true galaxies. Based on this simulated dataset, we use simple machine learning algorithms like Support Vector Machine (SVM), Random Forest (RF), or Multi-Layer Perceptron (MLP). This simple method is an asset for the comprehension of the classification process compared to Convolutional Neural Networks (CNN) that are commonly used. In this exploratory work, we found that 91.8\% of the simulated lenses are reported and classified with a precision of 95.6\%. This work reports thus promising results using MLP that are lower but rather close to the performances of a CNN. This work can be improved to reach comparable or even better results than today's state-of-the-art algorithms by studying residual image resulting from the substraction of a light profile to the original image.