Laboratório de Pesquisas Interdisciplinares em Informação Multimídia
IVAL: Automatic Visual Inspection System Applied to the Detection of Defects in Rolled Steels

This project addresses the development of an automated visual inspection system for rolled steel defects detection, by using Computer Vision techniques and Artificial Neural Networks. Unlike most common techniques, which are frequently based on manual estimations that lead to significant time and financial constraints, it presents an automatic system based on image analysis techniques for classification of defects with well-defined geometric shapes and the combination of a feature extraction technique with an Artificial Neural Network model for classification of defects with geometric shapes more complex. In this work, images extracted from real-world video streams realized in a rolled mill line of ArcelorMittal Timóteo industry were considered for detecting six types of defects: rolled welding, clamp, identification hole, exfoliation, oxidation and wave-form mark. For the proposed system, two well-known feature extraction techniques: (1) Principal Component Analysis (PCA) and (2) Haralick Descriptors as well as two Artificial Neural Network models: (1) Multilayer Perceptron (MLP) and a (2) Self-Organizing Maps (SOM) were evaluated. The system was successfully validated achieving overall classification accuracy of 87 percent and demonstrating its high potential to be applied in real scenarios.

See also

Prof. Flávio Cardeal. Fore more information: CV Lattes.

Luiz Martins. Fore more information: CV Lattes.

Prof. Paulo Almeida. Fore more information: CV Lattes.

© 2016 Piim-Lab - CEFET-MG - Av. Amazonas, 7675 - Nova Gameleira - Belo Horizonte - MG - Brasil | Tel: 55 31 3319-6870