Abstract: A classification method for corroded steel reinforcements based on the residual network(ResNet) was proposed， and a new scheme and technical reference for further realizing accurate quantitative evaluation on corrosion ratio of steel reinforcement on-site was provided. A total of 1478 steel reinforcements with diameters of 12mm and 14mm and corrosion ratio of 1.45%~56.10% are taken as the experimental objects. The images of steel reinforcements are taken by industrial cameras under laboratory conditions， and the total of 23648 sample images are acquired by using the data enhancement techniques. Then， 11 kinds of sample image labels are determined according to the corrosion ratios. The ResNet structure is built based on the deep convolutional neural network， and the Adam algorithm is used for iterative optimization. The classification accuracy and training rounds are evaluated by comparing the experimental results of different data sets. To verify the applicability of the proposed method， 6 data sets are formed from sample images of different diameter of steel reinforcements for the training and testing. The experimental research shows that after 100 rounds of iterative training， the classification accuracy of corrosion degree of steel reinforcements for 6 kinds of data sets are more than 93.2%， and the highest is 98.8%. It also indicates that the proposed method supports the high-precision classification of corrosion reinforcement with mixed diameters， and has good practical applicability.

Key words: corroded steel reinforcement; corrosion ratio; deep convolutional neural network(CNN); image classification; residual network(ResNet)