Labelling and Segmentation of Graphene Characterisation Images

We develop a cloud platform and workflow characterising graphene and functionalised graphene surfaces based on images obtained from transmission electron microscopy (TEM) and scanning electron microscopy (SEM) including image data management and processing following by feature extraction based on ML/AI techniques including high-performance segmentation and pre-annotation. Within that activity, we focus on two research questions related to detection strategies (RQ1) and segmentation performance (RQ2):

• RQ1: Manual preparation of segmentation masks (GFM, carbon holes, undefined) is time-consuming and prone to inconsistency. The study aims to evaluate whether foundation models (SAM2, zero-shot) or classical detection frameworks (Detectron2, Mask R-CNN with fine-tuning) can provide useful pre-annotations for TEM/SEM images.
• RQ2: Developing robust segmentation models for complex graphene images is particularly challenging due to the very limited availability of labeled data.

This study will explore strategies to achieve acceptable segmentation performance under such constraints.