Image recognition‐driven efficient detection method for ceramic powder hollow ratio: Optical mapping and deep learning

Abstract Efficient, nondestructive, and cost‐effective detection of the hollow ratio in ceramic powder remains an urgent technical challenge in the industrial field. In this study, a novel image recognition‐based method for detecting the hollow ratio of ceramic powder was proposed. Using 8 mol% yttrium‐stabilized zirconia powder as the research subject, a mapping relationship between the physical structure of hollow particles and their optical imaging characteristics was established. By conducting targeted image enhancement on the characteristics of ceramic particles, a high‐resolution image annotation dataset was constructed, and the detection performance of the YOLOv5s, YOLOv8s, YOLOv9s, and YOLOv11s models was systematically compared. The results indicate that the YOLOv11s model demonstrates superior overall detection performance. Building upon this finding and integrating the OpenCV scale calibration algorithm, automatic statistical analysis of particle size distribution (D10, D50, D90) and hollow ratio was achieved, significantly enhancing both efficiency and accuracy. This study highlights the promising application potential of the proposed method in the field of powder metallurgy.