Strip Steel Surface Defect Recognition Based on Novel Feature Extraction and Enhanced Least Squares Twin Support Vector Machine

Feature extraction and classification are two important steps in the process of strip steel surface defect recognition. Traditional methods of defect feature extraction are not of scale and rotation invariance. Moreover, traditional methods of defect classification have a conflict between efficiency and accuracy in. In order to solve the above two problems, a novel recognition method is proposed in this paper. On one hand, the novel defect feature extraction scheme is realized by building sampling benchmark scale (SBS) information for training dataset and using gradient magnitude and gradient orientation co-occurrence matrix (GMGOCM), gray level and gradient orientation co-occurrence matrix (GLGOCM), and moment invariant features. On the other hand, K-nearest neighbor and R-nearest neighbor algorithms are used to prune training dataset, and amplification factors of pruned samples are used to improve least squares twin support vector machine (LSTWSVM) classifier in efficiency and accuracy. The experimental results show that the novel recognition method can not only realize defect feature extraction with scale and rotation invariance but also realize defect classification with high efficiency and accuracy.