Image Super Resolution for Scanning Tunneling Microscopy and Atomic Force Microscopy

Scanning Tunneling Microscope (STM) and qPlus Atomic Force Microscopy (Q+AFM) image nano-material surfaces at atomic level, which have led to many major scientific breakthroughs. However, their resolutions are often limited by realistic experimental conditions. In this paper, we investigate image super resolution methods empowered by deep learning to go beyond STM and Q+AFM experimental limits. STM-SR and AFM-SR, adopting the Super-Resolution Generative Adversarial Networks (SRGAN) architecture, are developed and trained on high-resolution STM and Q+AFM datasets to convert low-resolution images into high resolution ones, respectively. Our results show that STM-SR and AFM-SR lead to significant resolution improvement compared to bicubic interpolation. This breaks the experimental barrier and reaches the resolution level near ideal conditions. This method can be applied to the study of large, complex molecules, which requires stringent experimental conditions.