This paper presents a machine learning (ML)-based method to enhance the spatial resolution of X-ray fluorescence (XRF) microscopy by decoupling the X-ray probe's impact from the XRF signal. A residual dense network (RDN) model is trained to transform low-resolution (LR) XRF images into super-resolved images, improving resolution for both simulated and experimental data. The method shows superior performance compared to conventional scanning XRF and is demonstrated on LiNi0.6Mn0.2Co0.2O2 (NMC) particles, significantly improving XRF tomography reconstructions.

Longlong Wu, Seongmin Bak, Youngho Shin, Yong S. Chu, Shinjae Yoo, Ian K. Robinson, Xiaojing Huang