A New Ensemble Approach based on Deep Convolutional Neural Networks for Steel Surface Defect classification

摘要

Steel surface defect recognition is a crucial component of automated steel surface inspection system which influences the quality of steel greatly. To improve the accuracy rate, an ensemble approach that integrating different deep convolutional neural networks (DCNNs) is proposed in this paper. Firstly, three different DCNNs are trained respectively with data augmentation to reduce over-fitting. Various optimization methods and tricks are used to reduce the error in the training procedure. Secondly, three well-trained models are combined. The experimental results show that the proposed approach made a state-of-art performance on accuracy rate and robustness in steel surface defect classification.