A fast method of steel surface defect detection using decision trees applied to LBP based features

Surface defect detection plays a significant role in quality enhancement in steel manufacturing. Support Vector Machines and neural networks are the most popular classifiers in this application. Decision trees are also known as other classifiers for steel defect detection yielding a fast but moderate performance. In this paper, we introduce a more accurate classification method by using decision trees and applying Principal Component Analysis (PCA) and Bootstrap Aggregating (Bagging) on features being extracted by a local binary pattern based operator. This methodology yields an enhanced accuracy and reinstates decision trees as fast and accurate classifiers for two-class classification of steel surface defects. In order to have a complete classification in a real-time automatic surface inspection, a multiclass Support Vector Machine should be cascaded to the decision tree classifier. The proposed classification system is considerably faster than the traditional schemes.