Microstructural Analysis of Transmission Electron Microscope Images of YBCO Superconducting Thin Films Using Machine Learning Image Recognition

In the Japan JST-CREST program, we aimed to establish a robust manufacturing process for high-temperature superconducting-coated conductors that ensures high performance regardless of equipment or operator variability. While process optimization is essential, understanding the microstructural origins of the superconducting properties is critical. It is empirically known that the magnetic field dependence of the critical current density (JC–Bcharacteristics) in YBa2Cu3Oy(YBCO) superconducting thin films correlates with their microstructure, particularly when doped with BaMO₃ (M = Zr, Sn, Hf), which forms nanorods acting as strong flux pinning centers. In this study, we focused on BaHfO₃ (BHO)-doped YBCO superconducting thin films and quantitatively evaluated impurities in BHO nanorods from transmission electron microscope images using the deep learning-based object detection algorithm YOLOv5. Through an investigation of the detected BHO nanorods, the matching field and BHO content are analyzed, and machine learning-based microstructural evaluation is demonstrated to have robust performance and high efficiency, making it effective for optimizing superconducting properties.