Development of a CNN for the Identification of Low Surface Brightness Galaxies

The detection and classification of low surface brightness (LSB) galaxies plays a critical role in understanding the formation and evolution of the universe. However, traditional observational techniques often face challenges in identifying these faint objects due to their low signal-to-noise ratio and their frequent overlap with brighter astronomicalsources. In this thesis, we present the development and implementation of a Convolutional Neural Network (CNN) made for the detection of LSB galaxies in astronomical images.This work is divided into two parts. The first part provides a theoretical foundation, introducing the concept of LSB galaxies and reviewing neural networks, with a specific focus on CNNs and their role in image processing tasks.The second part transitions into a practical framework, presenting PyTorch as the primary tool for the development and training of deep learning models. After exploring some example applications of CNNs using PyTorch, the design and implementation ofthe CNN made for the detection of LSB galaxies are detailed. The proposed model is trained on a synthetic dataset of galaxy images based on Euclid data, augmented with realistic noise and stars.Extensive testing on simulated datasets demonstrates the model’s ability to accurately recognize faint galaxies with high precision and generalization, outperforming traditional methods. The CNN achieves an 84.5% success rate in identifying galaxies with integrated magnitude values between 22 and 24, with even better results for higher luminosities. However, performance drops drastically for magnitudes between 24 and 26, which is the exact range where background noise becomes dominant.In summary, this work highlights the potential of CNNs as a tool in the field of astrophysics, enabling a more effective search of astronomical object and contributing to a deeper understanding of the cosmos.