ETDMS: Efficient two-stage diffusion model for accelerated SEM image super-resolution

Abstract

The scanning electron microscope (SEM) is a crucial tool for characterizing material microstructures, and it is renowned for its high resolution and depth of field. However, SEM image quality is affected by the scanning speed and resolution settings. When using SEM to capture fast-changing dynamic processes and other specific tasks, maintaining high image quality while using fast scanning mode is often tricky. To address these challenges, this paper introduces introduce an efficient two-stage diffusion model for accelerated SEM image super-resolution named ETDMS. The image denoising and super-resolution were divided into independent tasks based on SEM imaging principles and completed in two separate stages. Specifically, in Stage 2, a conditional lightweight encoder–decoder architecture SR network is proposed to replace the large U-Net in the traditional diffusion model and combine it with accelerated sampling technology to improve image generation efficiency. Experimental results prove that compared with previous super-resolution methods, the images generated by ETDMS significantly improve evaluation parameters, subjective visual quality, and detail generation.