Bayesian Networks-Based Defects Classes Discrimination in Weld Radiographic Images
Aicha Baya Goumeidane1, Abdessalem Bouzaieni2, Nafaa Nacereddine1, and Salvatore Tabbone2
1 Welding and NDT Research Center (CSC), BP 64, Cheraga, Algiers, Algeria [email protected]
2 Université de Lorraine, Loria, Vandoeuvre les nancy, France
Abstract. Bayesian (also called Belief) Networks (BN) is a powerful knowledge representation and reasoning mechanism. Based on probability theory involving a graphical structure and random variables, BN is widely used for classification tasks and in this paper, BN is used as a class discrimination tool for a set of weld defects radiographic images using suitable attributes based on invariant geometric descriptors. Tests are performed on a database of few hundred elements where the results are outstanding and very promising, since they outperform those given by powerful SVM classifiers.
Keywords: Bayesian networks · Weld defects · Geometric descriptors · Radiography Lecture Notes In Computer Science (LNCS) 9257, pp. 556–567, 2015. Springer Publishing