Optimization of the recognition of defects in flat steel products with the cost matrices theory

There is an Iurgent need for automatic systems for quality control of flat steel products to meet the ever increasing demand from customers, and for process control to prevent damage to production. To achieve these goals, those inspection systems should detect and recognize defects using image processing and pattern recognition methods. Such complex systems are difficult to compare and the specificity of each application is generally not taken into account. This paper describes a method for optimizing a system using a cost matrix approach. It produces an image which can be used to appreciate the quality of the classifier visually. We have applied the method in order to optimize the feature selection and to compare the classification methods used in a system designed for recognizing defects in flat steel products in a cold rolling mill. The results show significant improvements in the performance of the recognition system.