Metallographic Image Augmentation of Heat-resistant Steel Based on Improved DCGAN

abstract-Heat-resistant steel is an important material for the manufacture of utility boiler. Heat-resistant steel is subject to different grades of aging during long-term service, resulting in reduced material performance. The grade of aging and damage of heat-resistant steel depends on metallographic examination. At present, many scholars study automatic classification of heat-resistant steel aging grade based on convolutional neural network. However, the lack of metallographic image samples during module training is a common problem that seriously restricts the accuracy of aging grade classification model. Therefore, the method based on improved DCGAN network to realize metallographic image augmentation of heat-resistant steel with different aging grades is proposed in this paper. Experiments show that the accuracy of aging grade of metallographic images generated based on the proposed method is more than $98 \%$, which can provide sufficient metallographic images to support subsequent training and optimization of aging grade classification model of heat-resistant steel.