Copper Strip Surface Defects Inspection Based on SVM-RBF

Abstract

Recently, it becomes more important to ensure the quality of the products as copper strip manufacturing has been highly developed. The most difficult problem in process control and automatic inspection is classification of surface defects, so we develop an improved RBF (radial basis function) neural network classifier based on SVM (support vector machine) to automatically learn complicated defect patterns and use pseudo Zernike moment invariant as the defect feature. The optimal initial parameters of RBF network are gained through SVM, which has resolved the problems in traditional methods, e.g. long learning time, and easily getting into local minimum, etc. Furthermore, a BP learning algorithm is presented to adjust these hidden node parameters as well as the weights of the SVM-RBF. The experimental results show that the method is effective.