AI tools for plasma diagnostics by X-ray imaging and spectroscopy in the PANDORA project frame

Magnetized plasmas in compact traps offer a unique environment for fundamental research. PANDORA (Plasma for Astrophysics Nuclear Decay Observations and Radiation for Archeometry) is a multidisciplinary project focused on studying β decays in plasmas, using a novel facility that replicates stellar-like conditions. The project also supports applications to materials science, accelerator and ion source technologies, etc. A plasma diagnostics system based on a soft X-ray pinhole camera has been designed and implemented, with an innovative algorithm for Single-Photon Counting (SPhC) and High Dynamical Range (HDR) analysis. This enables space-resolved X-ray spectroscopy and the determination of magneto-plasma properties like local thermodynamic parameters (in terms of electron density and temperature) and confinement dynamics. This work presents results from an AI-based model in MATLAB designed to optimize the above mentioned algorithm. Using K-means clustering, events with similar features were grouped to identify those distinguishing real from spurious ones. A labeled dataset then is used to train a neural network to minimize pile-up, accelerating the recovery of high-resolution spectra and improving soft X-ray emission analysis. This contribution details the current neural network development stage and first applications to experimental data acquired during an experimental campaign carried out at the ATOMKI Laboratory.