Superpixel-based fuzzy clustering for the coating segmentation and thickness measurement of diverse coated fuel particles using local statistical features

Abstract

In Generation IV nuclear power systems, nuclear fuel performance and plant safety are significantly influenced by the coating thickness of coated fuel particle. However, coated fuel particles have a small enclosed spherical structure and diverse coating configurations, which makes significant challenge for the existing measurement methods. To address this problem, a superpixel-based fuzzy c-means clustering is proposed using colour, texture and position features (SFCM-CTP), for the coating segmentation and thickness measurement of diverse coated fuel particles. Initially, a morphologically-based central particle extraction method is developed to eliminate background interference from neighboring particles. Subsequently, an efficient particle image feature extraction method is proposed, which considers local statistical information, including colour, texture and position features in a comprehensive manner. Based on these features, an effective unsupervised coating segmentation method is proposed by combining simple linear iterative clustering (SLIC) and fuzzy clustering. The experimental results on the constructed particle dataset demonstrate that the proposed method not only performs well in the coating segmentation performance and thickness measurement, but also maintains high accuracy for particles with diverse coating configurations. The Dice values achieve 0.9724, 0.9742, 0.9333 on three configurations of particles, respectively.