Leveraging unlabeled SEM datasets with self-supervised learning for enhanced particle segmentation

2025/09/22 by Luca Rettenberger, Nathan J. Szymanski, Andrea Giunto +5 · 1 voice
Computer Science · Materials Science · #Domain Adaptation and Few-Shot Learning #Advanced Neural Network Applications #Machine Learning in Materials Science

paper · pdf · doi:10.1038/s41524-025-01802-3

openalex publication_date 2025/09/22 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/01

Abstract

Scanning Electron Microscopes (SEMs) are widely used in experimental science laboratories, often requiring cumbersome and repetitive user analysis. Automating SEM image analysis processes is highly desirable to address this challenge. In particle sample analysis, Machine Learning (ML) has emerged as the most effective approach for particle segmentation. However, the time-intensive process of manually annotating thousands of SEM images limits the applicability of supervised learning approaches. Self-Supervised Learning (SSL) offers a promising alternative by enabling knowledge extraction from raw, unlabeled data. This study presents a framework for evaluating SSL techniques in SEM image analysis, focusing on novel methods leveraging the ConvNeXtV2 architecture for particle detection. A dataset comprising 25,000 SEM images is curated to benchmark these proposed SSL methods. The results demonstrate that ConvNeXtV2 models, with varying parameter counts, consistently outperform other techniques in particle detection across different length scales, achieving up to a 34% reduction in relative error compared to established SSL methods. Furthermore, an ablation study explores the relationship between dataset size and SSL performance, providing actionable insights for practitioners regarding model selection and resource efficiency. This research advances the integration of SSL into autonomous analysis pipelines and supports its application in accelerating materials science discovery.

Citations

• ImageNet Large Scale Visual Recognition Challenge
• Momentum Contrast for Unsupervised Visual Representation Learning
• Swin Transformer: Hierarchical Vision Transformer using Shifted Windows
• Recent advances and applications of deep learning methods in materials science
• A ConvNet for the 2020s
• An autonomous laboratory for the accelerated synthesis of inorganic materials
• Segment Anything
• Enhancing performance of vision transformers on small datasets through local inductive bias incorporation
• Micrograph segmentations for DDEVD
• A Comprehensive Study of Vision Transformers on Dense Prediction Tasks
• A ConvNet for the 2020s
• Masked Autoencoders Are Scalable Vision Learners
• Barlow Twins: Self-Supervised Learning via Redundancy Reduction
• Transformers in Vision: A Survey
• Dense Contrastive Learning for Self-Supervised Visual Pre-Training
• An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
• Momentum Contrast for Unsupervised Visual Representation Learning
• 4D Spatio-Temporal ConvNets: Minkowski Convolutional Neural Networks
• Deep Residual Learning for Image Recognition
• ImageNet Large Scale Visual Recognition Challenge
• Swin Transformer: Hierarchical Vision Transformer using Shifted Windows

Discussions

• Can machines learn microscopy without labels? Work with KIT/UCB on self-supervised ConvNeXtV2 achieves ~41% error reduction over untrained models (15% vs ImageNet) for particle segmentation using 25k [bsky, 0 points, 0 comments]