Euclid Computer Vision

Project description

The Euclid space telescope from the European Space Agency (ESA) observes large areas of the sky with the goal of mapping the structure of the Universe, such as dark matter and dark energy. Within this data, strong gravitational lenses appear as distortions in galaxy images. Detecting and interpreting these systems is difficult, not only because they are rare, but also because their appearance depends strongly on how the data is processed and represented.

Euclid Computer Vision focuses on understanding these effects. The project explores how astronomical image data, color construction, scaling choices, and deep learning techniques influence both visual inspection and computational analysis. By combining data science, machine learning, image processing, and computer vision techniques, the project studies how different representations of Euclid data change what information becomes visible and usable.

Context

The project takes place in collaboration with the Euclid Strong Lensing Working Group (SWG), which forms part of the wider Euclid Consortium coordinated by the European Space Agency (ESA).

The project combines technical implementation with scientific exploration. With a main focus on data science and deep learning (ML) techniques. Its outcomes are intended to provide both practical tools and research insights for members of the Euclid Consortium, supporting future data releases and model development within the Strong Lensing Discovery Engine (SLDE).

Results

The project is currently positioned between TRL 3 and 4. This semester, the focus was on building on last semester’s research and moving toward a more integrated Euclid Computer Vision proof of concept. The most important outcomes are:

Integrated software workflow:
A proof of concept that connects data preprocessing, model inference, backend communication, and frontend development into one broader software setup.

Supervised classification:
Development and integration of machine learning classifiers for strong gravitational lens detection, starting with specific lens classes such as Einstein rings.

Unsupervised exploration:
Experiments with grouping and clustering techniques to explore patterns within lens candidate data and support future subclass analysis.

Data generation and transformation:
Continued work on generated and transformed data to support training, testing, and comparison between different image representations.

Image processing and representation:
Further experiments with preprocessing, scaling, colour construction, and transformation techniques to study how these choices influence visibility and classification.

Frontend and backend progress:
Development of supporting software components that make the project easier to use, extend, and connect to a future analysis interface.

These results show that the main concepts are technically feasible and have been tested in a limited software environment. However, more validation, larger datasets, and stronger integration are still needed before the system can be considered production-ready.

About the project group

A mixed group with students from different semesters and backgrounds within ICT. This project is very complex in data science and ML related topics, but even though this constraint, every person in the group managed to excel. This has mostly been because of the support within the group and communication between members. We worked around 2 to 3 days a week on this project for one whole semester.