基于卷积神经网络的遗传算法密码分析适应度函数

Fitness Function Based on Convolutional Neural Network of Genetic Algorithm for Cryptanalysis

遗传算法是密码分析的一种有效方法。遗传算法中适应度函数至关重要,直接影响到遗传算法的收敛速度以及能否找到最优解。在基于遗传算法的密码分析中,适应度函数主要基于语言字符频率统计特征。由于n-gram的统计字符仅包括26个英语字母,当密码字符集包含特殊字符或数字时,该方法效果不佳。提出一种新的基于卷积神经网络的遗传算法适应度函数,其基于卷积神经网络用于文本分类的原理。实验证明所提出的方法对经典密码分析是一种有效的方法。

Genetic algorithm is an effective method for cryptanalysis.Fitness function plays a very important role in genetic algorithm,which directly affects the convergence speed of genetic algorithm and whether the optimal solution can be obtained.In the cryptanalysis using a genetic algorithm,the fitness function to evaluate the quality of the key mainly relies on the statistical characteristics of the frequency of language characters.Since the statistical characters of n-gram only include 26 English letters,when the password character set contains special characters or numbers,the effect of this method is poor.This paper proposes a new genetic algorithm fitness function based on convolutional neural network(CNN),which is based on the principle of CNN for text classification.Experiments prove that the proposed method is an effective method for classical cipher cryptanalysis.