Researchers at the Hong Kong University of Science and Technology (HKUST) have created GrainBot, an AI toolkit designed to automatically extract and quantify microstructural features from microscopic images, to address long-standing challenges in materials science and engineering. The new system has the potential to accelerate the discovery and development of next-generation materials by providing a systematic method for converting complex image information into quantitative data and overcoming the limitations of current approaches that often focus on identifying simple features. GrainBot utilizes convolutional neural networks for accurate particle segmentation, integrating segmentation, feature measurements, and correlation analysis, alongside custom algorithms that measure parameters such as particle surface area and groove shape. “This study highlights the broader relevance of emerging AI-driven scientific infrastructures,” said GUO Yike, professor and chair of the School of Computer Science and Engineering at the University of Hong Kong and co-author of the study, suggesting a future where data-driven materials research will be more streamlined and insightful.
GrainBot toolkit automates microstructure quantification from microscopic images
A team led by Professor ZHOU Yuanyuan designed GrainBot to go beyond approaches limited to simple feature identification and image classification to provide an integrated solution that includes segmentation, feature measurement, and correlation analysis. Tests using metal halide perovskite thin films, essential for high-efficiency solar cells, have created a database of thousands of annotated particles, revealing relationships between characteristics such as particle size, groove shape, and surface roughness that were previously difficult to quantify. Beyond statistical analysis, we used interpretable machine learning models to determine how these features influence each other. The researchers were able to examine how parameters such as grain surface area and grain boundary groove angle work together to shape the depth of surface depressions.
Professor GUO Yike said, “GrainBot shows how AI can transform complex microscopy images into structured, reproducible datasets that can be easily shared, reanalyzed, and integrated into larger research platforms.” Professor Zhou emphasizes the toolkit’s accessibility, saying, “Our goal is to lower the barriers to integrating microscopy characterization into data-driven research and autonomous laboratory platforms.” The study, published in Matter on February 26, 2026, provides a strategic framework that can be applied to other polycrystalline thin films and will investigate the correlation between microstructure and device stability.
Segment particles and measure surface shape with convolutional neural networks
Traditionally, microstructural quantification has relied on manual analysis of microscopic images, a process that is time-consuming and prone to discrepancies, limiting the ability to fully understand structure-property relationships. GrainBot provides a systematic solution, converting visual data into quantitative descriptors and enabling the creation of large-scale standardized databases. The core functionality of this toolkit centers around accurate particle segmentation achieved through convolutional neural networks and algorithms designed to measure specific geometric features such as particle surface area and grain boundary shape. This allows researchers to go beyond qualitative observations and establish statistically significant correlations between microstructural parameters, as demonstrated through application to metal halide perovskite thin films, an essential material for high-efficiency solar cells. Professor Zhou emphasized the toolkit’s accessibility and said its goal is to lower the barriers to integrating microscopy characterization into data-driven research and autonomous laboratory platforms.
As scientific workflows become more automated and data-intensive, such toolkits will serve as the primary engine of future autonomous laboratories, continuously feeding standardized microstructural metrics into decision-making systems for materials discovery and optimization.
GUO Yike, Professor and Professor, School of Computer Science and Engineering and School of Electrical and Computer Engineering, University of Hong Kong
Verification of relationship analysis between microstructure and properties of perovskite thin films
The team, led by Professor ZHOU Yuanyuan, Associate Professor in the Department of Chemical and Biological Engineering at HKUST, leveraged GrainBot to build a comprehensive database containing thousands of individual particles. Each particle is meticulously annotated with multiple microstructural parameters obtained from atomic force microscopy (AFM) images. The validation process revealed common distribution patterns and enabled statistical analysis to reveal how these parameters interact to influence material properties. Professor Zhou added that the toolkit is aimed at supporting researchers who require a consistent quantitative description of microstructures. This systematic approach to understanding grain morphology, such as grooves, depressions, and ridges at grain boundaries, is particularly important for increasing the long-term stability of perovskite solar cells, according to research published in the journal Matter.
Our goal is to lower the barriers to integrating microscopy characterization into data-driven research and autonomous laboratory platforms.
