Automated SEM-based nanoparticle metrology for materials characterization via segmentation and robust scale-bar recognition.

Quantitative nanoparticle metrology is essential for establishing structure-property relationships in catalysis, drug delivery, optoelectronic materials, battery materials, and synthesis optimization. Scanning electron microscopy (SEM) provides direct access to particle morphology, but high-throughput and reproducible analysis remains limited by ambiguous particle boundaries, particle agglomeration, and unreliable scale calibration. Here, we report an automated SEM-based nanoparticle metrology workflow that converts raw SEM images into scale-calibrated particle-size distributions and morphology statistics. The framework integrates Multi-scale U-shaped Kolmogorov-Arnold Network (MU-KAN) segmentation with YOLOv11-based scale-bar localization and optical character recognition-assisted scale annotation parsing. On the public NanoSEM-464 and NanoSEM-1707 benchmarks, MU-KAN achieves IoU values of 0.9296 and 0.8243 and F1-scores of 0.9630 and 0.9029, respectively. The scale recovery module gives a mean relative error of 3.8604% on a 370-image SEM scale-bar dataset. The resulting workflow provides particle area, equivalent circular diameter, circularity, and complementary shape descriptors, enabling automated, reproducible, and physically meaningful nanoparticle characterization for materials research.