Mission
Slitless spectroscopy is a powerful observational tool in astronomy, enabling the simultaneous collection of spectra from numerous objects. It is widely used in space telescopes like the HST, the JWST, Euclid and the forthcoming Roman Space Telescope for large-scale surveys. However, this technique presents numerous challenges, particularly with the management of overlapping spectra (spectroscopic contaminants) and the extraction of precise redshift measurements from low-resolution and noisy data.
In recent years, deep learning (DL) has revolutionized many areas of data analysis, especially for tasks involving pattern recognition and noise reduction. This proposal explores the application of DL models to improve the analysis of slitless spectroscopic data, focusing on 1. the identification and subtraction of contaminant spectra, 2. redshift determination directly from spectrograms, potentially associated to external multi-band photometric observations.
The aim of this research is to develop DL models that address two key challenges in the context of the Euclid survey:
- Spectroscopic contaminant identification and subtraction: Creating a DL-based method to accurately identify and remove overlapping spectrograms from target one.
- Redshift estimation from joint analysis of spectrogram and multi-band photometry: Combining low-resolution spectrograms with multi-band photometric data (colors and morphology) to boost redshift prediction accuracy and reliability.
Background and Motivation
Slitless spectroscopy allows for efficient, wide-field surveys, but its inherent challenges limit its full potential. Spectroscopic contamination occurs when the spectra of nearby objects overlap, which requires careful modeling and ad-hoc correction. Similarly, redshift determination in slitless data can be complicated by the lack of well-identified features.
Traditional methods rely on empirical correction for contamination and use template fitting or line identification for redshift estimation on extracted 1D spectra. These methods are sub-optimal, especially for faint extended objects such as distant galaxies.
Recent research shows DL’s potential for pattern recognition and noise reduction in astronomy. Moreover, the integration of multi-band photometric data, probing the galaxy morphology and color, can provide crucial
supplementary information to improve redshift estimates. By combining spectral and photometric data, the model can leverage the advantages of both data types, enhancing the accuracy of cosmological measurements.
Methodology
- Identification and Subtraction of Spectroscopic Contaminants: a convolutional neural network (CNN)-based architecture will be designed to detect and subtract contaminating spectra. The model will be trained using supervised learning, with labeled training data indicating which spectrogram belong to the target source and which are contaminants.
- Redshift Estimation including Multi-band Photometric Data: multi-band photometric images are available (VIS, NIR, EXT) and will be used to supplement slitless spectroscopic data (SIR). A hybrid DL model will be built to integrate spectroscopic data and multi-band photometric features. A CNN will handle spectroscopic data, while a separate neural network will process the photometric features. The two networks will be merged into a combined architecture that uses both data sources for redshift estimation.
Conclusion
This research will deliver significant advancements in slitless spectroscopic analysis through:
- Automated contaminant removal: A DL model that efficiently identifies and removes spectroscopic contaminants, improving data quality.
- Improved redshift estimation: A hybrid DL model that combines spectroscopic data with multi-band photometric information, leading to more accurate redshift measurements, particularly for faint or high-redshift objects.
=================
For more Information about the topics and the co-financial partner (found by the lab !); contact Directeur de thèse - y.copin@ipnl.in2p3.fr
Then, prepare a resume, a recent transcript and a reference letter from your M2 supervisor/ engineering school director and you will be ready to apply online before March 14th, 2025 Midnight Paris time !