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

Area of Science:

• Materials Science
• Microscopy
• Image Processing

Background:

• Scanning Electron Microscopy (SEM) is vital for material microstructure analysis.
• SEM image quality degrades with faster scanning speeds, hindering dynamic process observation.
• Existing super-resolution methods struggle with high-speed SEM imaging challenges.

Purpose of the Study:

• To develop an efficient super-resolution method for SEM images captured at high speeds.
• To improve the quality and detail of SEM images without compromising acquisition speed.
• To address the trade-off between scanning speed and image fidelity in SEM.

Main Methods:

• Introduced a novel two-stage diffusion model named ETDMS for accelerated SEM image super-resolution.
• Separated image denoising and super-resolution into distinct stages based on SEM principles.
• Proposed a conditional lightweight encoder-decoder super-resolution network in Stage 2, replacing traditional large U-Nets and incorporating accelerated sampling.

Main Results:

• ETDMS significantly improved SEM image quality compared to existing super-resolution techniques.
• Enhanced evaluation parameters, subjective visual quality, and fine detail generation were observed.
• The model demonstrated effectiveness in accelerating image generation while maintaining high fidelity.

Conclusions:

• ETDMS offers an efficient solution for high-quality, accelerated SEM image super-resolution.
• The two-stage approach and lightweight network design improve generation efficiency.
• This method is beneficial for capturing fast-changing dynamic processes and specific SEM applications.