The first annotated set of scanning electron microscopy images for nanoscience

Area of Science:

• Materials Science
• Computer Science
• Data Science

Background:

• Scanning Electron Microscopy (SEM) generates high-resolution nanoscale images crucial for scientific research.
• The development of artificial intelligence (AI) for image recognition requires large, well-annotated datasets.
• Publicly available datasets for SEM images are scarce, hindering AI model training and validation.

Purpose of the Study:

• To present the first publicly available, human-annotated dataset of Scanning Electron Microscopy (SEM) images.
• To provide a comprehensive collection of nanoscale SEM images for AI-driven image recognition tasks.
• To establish a foundational dataset for advancing automated analysis of microscopic structures.

Main Methods:

• Collection and curation of approximately 26,000 SEM images at the nanoscale.
• Human annotation and classification of images into 10 distinct categories.
• Organization of images into 4 labeled training sets suitable for machine learning.
• Implementation of a preliminary hierarchical structure for image categories.

Main Results:

• A novel dataset of 26,000 human-annotated SEM images is now publicly available.
• The dataset covers diverse nanoscale features including particles, nanowires, films, patterned surfaces, MEMS devices, tips, and biological samples.
• Images are categorized into 10 classes with a hierarchical structure, facilitating varied recognition tasks.
• Four distinct labeled training sets have been generated for image recognition model development.

Conclusions:

• The release of this annotated SEM dataset significantly advances the field of AI-powered microscopy.
• This resource will accelerate the development and deployment of accurate image recognition algorithms for nanoscale analysis.
• The dataset's comprehensive nature and hierarchical structure offer broad applicability across scientific disciplines utilizing SEM.