基于流形学习能量数据预处理的模板攻击优化方法

As the key object in the process of template analysis, power traces have the characteristics of high dimension, less effective dimension and unaligned. Before effective preprocessing, template attack is difficult to work. Based on the characteristics of energy data, a global alignment method based on manifold learning is proposed to preserve the changing characteristics of power traces, and then the dimensionality of data is reduced by linear projection. The method is validated in Panda 2018 challenge1 standard datasets respectively. The experimental results show that the feature extraction effect of this method is superior over that of traditional PCA and LDA methods. Finally, the method of template analysis is used to recover the key, and the recovery success rates can reach 80% with only two traces.