In a new publication in Machine Learning: Science and Technology, we describe a multiscale Bayesian approach for quantifying and denoising EDX spectrum images. The approach is innovated by LSME postdoc Pau Torruella working in collaboration with Abderrahim Halimi from Heriot-Watt University. Building on previous LSME developments for EDXS data processing, this approach leverages both the spectral and spatial correlations of the datasets. It has particularly strong capabilities when working with low signal/high noise data, as we demonstrate with atomic resolution studies of ferroelectric thin films from our collaborators Céline Lichtensteiger and Ludovica Tovaglieri at the Department of Quantum Matter Physics at the University of Geneva. All resources are openly available online.
