Abstract
Observing chemical reactions in complex structures such as zeolites involves a major challenge in precisely capturing single-molecule behavior at ultra-high spatial resolutions. To address this, a sophisticated deep learning framework tailored has been developed for integrated Differential Phase Contrast Scanning Transmission Electron Microscopy (iDPC-STEM) imaging under low-dose conditions. The framework utilizes a denoising super-resolution model (Denoising Inference Variational Autoencoder Super-Resolution (DIVAESR)) to effectively mitigate shot noise and thereby obtain substantially clearer atomic-resolved iDPC-STEM images. It supports advanced single-molecule detection and analysis, such as conformation matching and elemental clustering, by incorporating object detection and Density Functional Theory (DFT) configurational matching for precise molecular analysis. the model's performance is demonstrated with a significant improvement in standard image quality evaluation metrics including Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM). The test conducted using synthetic datasets shows its robustness and extended applicability to real iDPC-STEM images, highlighting its potential in elucidating dynamic behaviors of single molecules in real space. This study lays a critical groundwork for the advancement of deep learning applications within electron microscopy, particularly in unraveling chemical dynamics through precise material characterization and analysis.