Strip surface defect recognition algorithm based on PCA and improved BP neural network

摘要

In order to meet real-time requirements of strip surface defect detection, the extracted 41 original features of strip surface defects are reduced-dimensionally optimized through the principal component analysis. As the study samples of improved BP neural network, the 10 integrated features are used to train and test network. The trained network model is saved for on-line recognition and classification. Experimental results show that feature space can be optimized by principal component analysis, and the network structure is simplified (network input neuron number is reduced to 10 from 41), not only is the network convergence rate accelerated, but also is the network recognition rate improved, thus the real-time requirements and accuracy of strip surface defect detection are met.