基于数据随机性特征和极速学习机的加密数据流识别

Encrypted Data Stream Recognition Based on Data Randomness and ELM

为了在不解密加密数据的前提下获取加密数据流的类型信息,提出一种基于数据随机性特征和模式识别的加密数据流识别方法。该方法利用加密数据与非加密数据,或者不同类型加密数据0,1分布的随机性特性作为分类特征,再利用模式识别方法对不同数据进行建模,从而实现对不同类型数据的自动识别。首先利用NIST随机性测试方法对数据流进行分析,将得到的15类随机性测试得分作为分类特征;然后对不同类型的数据流分别建立分类模型;最后利用训练好的数据模型对未知数据流进行识别。仿真实验显示,与仅用单个随机性特征进行明密数据识别相比,采用模式识别方法可以将错分率由原来的60%以上下降到30%左右;进一步利用滤波器方法对15类随机性特征进行优化降维,平均错分率进一步下降到15%左右。

This paper presented the identification method of encrypted data stream based on data randomness and pattern recognition without decrypting the encrypted data.The method uses the randomness distribution characteristics of different data as classification features,and then uses the pattern recognition method to classify different types of data.Firstly,the randomness test method NIST is used to analyze the data stream,getting the 15 kinds of randomness test values as the classification feature.Then the method creates classification models for different types of data streams.Finally,the method uses the trained model to identify the unknown data stream.Simulation results show that using the15 kinds of features,the proposed method can effectively classify the different types of data stream,and the error rate decreases from 60%to 30%.Using the feature optimization method,the error rate drops to 15%.