Unsupervised weld defect classification in radiographic images usingmultivariate generalized Gaussian mixture model with exactcomputation of mean and shape parameters

Type : Publication
Auteur(s) :  Nafaa Nacereddine, Aicha Baya Goumeidane, Djemel Ziou
Année :  2019
Domaine : Electronique
Revue : Computers in Industry
Résumé en PDF :  (résumé en pdf)
Fulltext en PDF :  (.pdf)
Mots clés :  Mixture model, Multivariate GGD, radiography, weld defect, classification

Résumé : 

In industry, the welding inspection is considered as a mandatory stage in the process of quality assurance/quality control. This inspection should satisfy the requirements of the standards and codes governing themanufacturing process in order to prevent unfair harm to the industrial plant in construction. For thispurpose, in this paper, a software specially conceived for computer-aided diagnosis in weld radiographictesting is presented, where a succession of operations of preprocessing, image segmentation, featureextraction andfinally defects classification is carried out on radiographic images. The last operationwhich is the main contribution in this paper consists in an unsupervised classifier based on afinitemixture model using the multivariate generalized Gaussian distribution (MGGD). This classifier is newlyapplied on a dataset of weld defect radiographic images. The parameters of the nonzero-mean MGGDbasedmixture model are estimated using the Expectation-Maximization algorithm where, exactcomputations of mean and shape parameters are originally provided. The weld defect database representfour weld defect types (crack, lack of penetration, porosity and solid inclusion) which are indexed by ashape geometric descriptor composed of geometric measures. An outstanding performance of theproposed mixture model, compared to the one using the multivariate Gaussian distribution, is shown,where the classification rate is improved by 3.2% for the whole database, to reach more than 96%. Theefficiency of the proposed classifier is mainly due to theflexiblefitting of the input data, thanks to theMGGD shape parameter.