In order to solve the difficult problem of traditional template analysis in practical attacks, this paper focused on the VGGNet network model with excellent feature extraction capabilities in the field of image recognition. It proposed a new template attack method based on VGGNet network model. In order to prevent the signal quality from affecting the accuracy of the model, it used the correlation power analysis method to test the quality of the collected side-channel signal. In order to adapt to the dimensional characteristics of the side-channel signal data, it made modest adjustments to the network model structure. In the process of network training, it focused on issues such as slower gradient descent, gradient disappearance, and overfitting, and used a five-fold cross validation method to validate the trained model. The final experimental results show that the test success rate based on the VGGNet model is 92.3%, which is 7.7% higher than the traditional template attack effect.

[2019-09-05] Printed Article /doc/2018.04.0255/v2