Classification based on multi-classifier of SVM fusion for steel strip surface defects

Aiming at the existing problems in pattern recognition of surface defect images of steel strips, a new classification and recognition method based on multi-classifier of support vector machine (SVM) fusion is proposed to solve them. Firstly extracted the Hu invariant moment features, gray features and texture features, and devised SVM classifiers based on the different features and combination features to classify the defects. Then, using the majority voting procedure fused the defect classification results of these single classifiers based on three different features, compared with the results of the classifier based on combination features, if it is equal then the classification results is gotten, otherwise correcting results with the binary classifier. Experimental results demonstrated the fused features and combined classifiers are the definite improvement over non-fused features and single classifier, the classification rate is up to 98%.