Affiliation:
1. International Laboratory for Insulation and Energy Efficiency Materials, College of Materials Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing, 211106, China
2. School of Computer Engineering, Jinling Institute of Technology, Nanjing, 211169, China
Abstract
C/SiC composite has excellent performance, but their damage patterns cannot be effectively detected, and it is tough to obtain a large amount of label data. Therefore, an autonomous damage image recognition based on transfer learning is proposed in this paper. A C/SiC composite damage
image data set is collected from experiments and papers, including fiber pullout, matrix cracking, interface debonding, and fiber breakage images. Pre-trained DenseNet121 was transferred to train and classify this data set. Pre-trained AlexNet and other four networks are selected for comparison.
The results of experiment illustrate that the accuracy of directly using the DenseNet121 model is 70.536%, and the accuracy of the transferring pre-trained AlexNet, VGG16, ResNet18, and MobileNet-v2 on the validation set is 96.429%, 84.598%, 94.420%, 94.866%, respectively. The accuracy in
this paper is 97.098%. As the number of damage images of C/SiC composite is limited, the recognition accuracy of the method is significantly improved.
Publisher
American Scientific Publishers
Subject
General Materials Science
Cited by
3 articles.
订阅此论文施引文献
订阅此论文施引文献，注册后可以免费订阅5篇论文的施引文献，订阅后可以查看论文全部施引文献