A method for assessing the distribution of uranium spheres in fuel elements using x-ray computed tomography

The paper presents a method for automated assessment of uranium spheres distribution within an object based on processing results from X-ray computed tomography. The task of object detection based on reconstructed cross-sections and possible solutions are analyzed. A combined approach is proposed, integrating YOLO deep learning method for initial detection and the DBSCAN algorithm for spatial data clustering. It is shown that using YOLO method in this case ensures high detection accuracy, while DBSCAN clustering allows elimination of artifacts caused by multiple detections of the same sphere on adjacent slices. Further improvements in distribution estimation quality are suggested through optimization of clustering parameters and integration of deep learning methods for handling low-contrast images.