This thesis at the interface between nuclear instrumentation and applied mathematics consists in developing and implementing advanced methods for processing spectral data from CdTe Caliste detectors for high-energy photons. These sensors, resulting from fundamental research in space astrophysics, are the basic building block of the Spid-X gamma camera born from joint technological developments between the CEA and the company 3D PLUS. It aims at characterizing radiative environments in the framework of nuclear surveillance, for the safety of nuclear operations or research facilities, or for the dismantling of installations.
The methods studied will use Deep Learning tools with the objective of analyzing gamma spectra acquired in a complex environment inducing spectral distortions, potentially difficult to interpret with classical algorithms.
For this purpose, the PhD student will carry out the following lines of study:
- The identification of radioelements and the measurement of their proportion in the signal with one or several absorbing and scattering materials between the sources and the detector (methods: Monte-Carlo Geant4 simulations, Bayesian neural networks, confidence robust learning and experimentation).
- Determination of the nature of the material crossed and the thickness crossed (methods: adversarial neural networks (GANs), self-encoding, experimentation).
- The application to coded mask imaging methods. Depending on the results obtained in the two previous axes and the resulting discovery space, the methods may be applied to the theme of coded mask methods for gamma-ray imaging.