Multi-class Classification Method Based on Support Vector Machine with Hyper-sphere for Steel surface Defects

In order to obtain better recognition result for steel surface defects, a multi-class classification method based on support vector machine with hyper-sphere is proposed in this paper. On one hand, support vector data description (SVDD) is used to obtain one-class classification hyper-sphere for each type of defect dataset. In order to ensure the representability of sparse boundary samples and overcome the adverse effect of noise, adjustable factor and parameter constraint are introduced. On the other hand, based on the boundary samples, a support vector machine with multiple hyper-spheres (MH-SVM) model is formulated. It not only absorbs the hyper-sphere of SVDD, but also uses one-class classification hyper-sphere to amend and restraint itself. The above two aspects ensure the classification performance of MH-SVM and reduce the sensitivity to noise. Experimental results show that the proposed multi-class classification method has high classification accuracy and efficiency for defect datasets, especially for corrupted datasets.