Abstract: Considering that the accuracy of classification in corroded rivets is low and manual inspection is the
main method，a Tree-CNN ( Convolutional Neural Networks) classification method is proposed． This method is
specially designed for classifying corroded rivets on aircrafts． In order to improve the classification accuracy of
Tree-CNN method，the structure of the tree is determined by the confusion matrix of rivet categories which is
calculated in normal CNN method． The depth of the tree is three for five-classification of corroded rivets．
Experimental results show that by using the Tree-CNN method，the accuracy of classifying corroded rivets can
reach up to 86． 5%，which is effective in classification in corroded rivets．

Key words: tree structure, convolutional neural networks( CNN) network, rivet classification, confusion matrix