Automatic Recognition of Steel Deterioration Grade By CNN

Regular observation of steel microstructure in met-allographic image allow us to know its deterioration grade and take appropriate measures to ensure safe operation of coal-fired power plant. To eliminate the influence of subjective factors in the inspection process, this paper used an improved ResNet50 combined with attention mechanisms and Pyramid Convolution for automatic recognition of steel deterioration grade. At same time, online label-smoothing is used to exploit the similarity of images in adjacent deterioration grade to improve recognition performance. We collected metallographic images from work-pieces in coal-fired power plant to validate our method. The experimental results show that the recognition accuracy reached 91.5%, which is 8.15% higher than ResNet50, and error cases occured only in adjacent grade. The proposed method can meet the requirements of automatic steel deterioration recognition task and is sufficient to support the safe operation of coal-fired power plant.