Weak-Edge Extraction of Nuclear Plate Fuel Neutron Images at Low Lining Degree

• Published in: Applied sciences Vol. 13; no. 8; p. 5090
• Main Authors: Wang, Qibiao, Luo, Yushi, Sun, Yong, Wu, Yang, Tang, Bin, Peng, Shuming, Tuo, Xianguo
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.04.2023
• Subjects: Accuracy; Algorithms; Aluminum; Cladding; Cold; Deep learning; Edge detection; Imaging; Localization; Methods; Morphology; neutron imaging; Neutrons; Noise; Noise control; Nondestructive testing; Nuclear cross sections; Nuclear energy; Nuclear fuel elements; Nuclear industry; Nuclear reactors; Pellets; phase congruency; plate fuel element; Wavelet transforms; Zirconium
• ISSN: 2076-3417; 2076-3417
• DOI: 10.3390/app13085090

Neutron imaging is an effective nondestructive testing (NDT) technique widely applied to detect structural defects and the enrichment of nuclear fuel elements due to its high penetration and nuclide-sensitive properties. Since the fuel element pellet is sealed in the cladding, the transmission imaging result is a superposition of the two parts. Therefore, the attenuation of neutrons by the cladding is interference that must be considered in the enrichment analysis. It is necessary to extract and separate cladding and pellets using an edge extraction method. However, the low neutron cross-section of the cladding material (e.g., aluminum and zirconium) leads to poor grayscale contrast at the cladding edge in the imaging result, and the intensity of the cladding edge is significantly lower than that of the pellet edge. In addition, affected by the noise from the imaging environment, the boundaries of targets are further blurred, making edge detection more challenging. Traditional detection algorithms extract the weak edges of cladding incompletely, and the results are discontinuous, with obvious edge breaks and missing areas. This paper proposes a method to extract edges in neutron images based on phase congruency (PC). This study utilized the classical perceptual field model to improve contrast at weak edges. The enriched edge map was generated using our PC model from six directions, allowing more weak edges to be detected accurately. The non-maximum suppression ensured precise localization and avoided edge breaks. Furthermore, the edge results were optimized by eliminating noise through morphological operations. The experimental results demonstrate that the proposed method effectively detects the weak edges of the cladding, is superior in accuracy and integrity to traditional detection, and is able to obtain stable and reliable results with different materials of neutron images. The edge integrity improved by 64.1%, and the edge localization accuracy reached 94.3%. The extracted edge information is useful in the next stage of the high-precision enrichment analysis.