Microstructure Recognition by Deep Learning

Abstract

Deep learning by convolution neural network (CNN) was applied to recognize a microstructure of steels. Three typical CNN-models such as LeNet5, AlexNet, and GoogLeNet were examined their accuracy of recognition. In addition to a model, an effect of learning rate, dropout ratio, and mean image subtraction on recognition accuracy were also investigated. Through this study, the potency of deep learning for microstructural classification is demonstrated.