The appearance quality of nuclear fuel pellets is closely associated with the operational safety and efficiency of nuclear reactors, and must exhibit high quality before being loaded into reactors. Currently, visual inspection by human operators is employed to control the quality of nuclear fuel pellets. However, manual inspection is highly subjective, prone to visual fatigue, carries risks of false and missed detections, and lacks the capability for quantitative defect analysis. To effectively detect surface defects of UO₂ nuclear fuel pellets, an automated inspection system based on machine vision has been designed. The system adopts a modular architecture: first, preprocessing the acquired pellet images; then, utilizing image segmentation techniques tailored to the surface characteristics of the pellets to extract defect regions; subsequently, applying connectivity analysis for defect examination; and finally, employing a combined approach of Support Vector Machine (SVM) classifier and Convolutional Neural Network (CNN) for defect classification, enabling automatic identification of pellet surface defects. The results demonstrate that the system achieves high detection accuracy and efficiency, meeting the requirements of industrial automated inspection.

nuclear fuel pellets  /  UO₂  /  defect analysis  /  automatic detection  /  machine vision  /  image segmentation  /  module