Research on strip surface defect classifier used RBF neural networks based on PCA

Aiming at the existing problems in pattern recognition of surface defect images of steel strips, a RBF neural network classification and recognition method based on principal component analysis (PCA) is proposed to solve them. Using PCA to extract the main characteristics of the sample data which computed by the image of strip surface defects to achieve the optimal sample characteristics data compression, thereby reducing the sample characteristics data dimension. The principal components are used as the input of neural network. The RBF neural networks center is automatic selected using the nearest neighbor clustering method. The simulation results show that comparing with general RBF neural networks, the RBF neural networks improved by the nearest neighbor clustering method have higher classification accuracy, and can simplify the network structure.