Research On Cryptosystem Identification Scheme Based On Machine Learning

With the continuous acceleration of the digitization of daily life,emerging technologies such as mobile payment,digital currency and blockchain have begun to rise,all of which are inseparable from the rapid development of modern cryptography.With the widespread use of cryptographic content,a large amount of encrypted data has been generated.Because of the particularity of encrypting data itself,the task of analyzing the cryptosystem used in the encrypting process,no matter from the perspective of cryptanalysis or data analysis,has become a major difficulty.Moreover,identifying the cryptosystem is an important prerequisite for further analysis of the ciphertext.Aiming at the problem of cryptosystem recognition,this paper studies the theoretical basis of recognition scheme,the method of extracting ciphertext features and the construction of classifier model.Regarding the theoretical basis of the recognition of cryptosystems,some studies have given a simple formal description.However,only the definition of the problem and scheme can not make substantive progress on the main difficulties such as extracting ciphertext features and building classification model.In view of this,on the basis of the definition system of the existing cryptosystem identification problem,this paper further characterizes the ciphertext features used in the cryptosystem identification problem.According to this model,different attributes of ciphertext characteristics are analyzed.And through a comparative experiment,the influence mechanism of different attributes of ciphertext characteristics on the recognition effect is studied.Analysis of experimental results reveals that among the ciphertext features used in the identification scheme of the cryptosystem,different features have different recognition effects on the same cryptosystem.According to the ciphertext feature model,the organization of the ciphertext data and the mathematical transformation method used to extract the features have the greatest impact on the recognition results of the cryptosystem identification scheme.Based on the experimental results,we redesigned the feature extraction method of ciphertext,and proposed a dynamic feature recognition scheme which can adapt to the recognition scenarios of multiple cryptosystems.This scheme combines Relief feature selection algorithm and heterogeneous integrated learning.First,the feature selection algorithm is used to select the character set that will be used when extracting features.According to different recognition scenarios,a character set suitable for the current recognition task is selected,and the new character set is used as the basis for extracting ciphertext features in the next stage.Then use the classifier to complete the recognition task.The experiment and analysis are carried out on the ciphertext data set composed of 10 encryption algorithms including stream cipher,block cipher and asymmetric cipher.In addition to selecting the entropy value and probability of the ciphertext as the method of extracting the features of the ciphertext,this paper refers to the related research results of document classification,and uses the doc2 vec technology to convert the ciphertext text into a vector to complete the recognition task.The experiment and analysis are carried out on the ciphertext data set composed of 9 encryption algorithms.The experimental results show that,in the recognition task including multiple types of cipher systems,the recognition scheme of extracting different ciphertext features under different specific recognition scenarios has better recognition effect and higher accuracy than the recognition scheme using a single ciphertext feature.