Overcoming Labelled Data Scarcity for Defect Classification in Scanning Tunneling Microscopy
Researchers have developed a novel approach to classify defects in scanning tunneling microscopy (STM) images by addressing the scarcity of labeled data. This method employs a combination of few-shot learning and unsupervised learning, eliminating the need for extensive manual annotation. The technique has demonstrated strong generalization capabilities, successfully identifying atomic features on various surfaces with minimal additional labeled data, paving the way for more efficient and material-agnostic STM image segmentation. AI
IMPACT Potential to accelerate materials science research by automating image analysis.