A Research Of Cryptographic Algorithm Recognition Scheme Based On Machine Learning

Intelligence analysis is an indispensable part in information confrontation activities.It is the main way to obtain useful information from intercepted unknown information.Cryptanalysis is the main subject of modern intelligence analysis activities.Identifying the cryptographic algorithm to which ciphertext belongs is a necessary prerequisite for cryptanalysis in reality.The study of effective methods for identifying cryptographic al-gorithm has important practical significance and application value.The research content of this paper is about the recognition of cryptographic algo-rithms based on deep learning,and specific research and analysis are carried out around two aspects of selecting ciphertext features and recognition algorithms.The main work and results of the thesis are as follows:(1)Randomness test is a common method for identifying cryptographic algorithms.This paper studies the random feature distribution of ciphertexts generated by various cryptographic algorithms.The conclusion obtained is that the random feature distribution of different cryptographic algorithms is also different in some items of randomness tests.The influence of the length of the ciphertext on the feature distribution is studied.Provide a necessary reference for the selection of randomness test items and the size of ciphertext segmentation length for cryptographic algorithm identification.(2)5 block cipher algorithms and 3 public key cipher algorithms are selected,and text files are used as plaintexts to generate ciphertexts with different cipher algorithms.The ciphertext-feature is designed based on the selected randomness test items,and then extract ciphertext-features of different cryptographic algorithms.The random forest algorithm is used to identify 8 kinds of cryptographic algorithms,and a high recognition accuracy rate is achieved.The effectiveness of the designed features in the identification task of cryptographic algorithms is verified.(3)Studied the application of deep learning algorithms in recognition tasks of cryp-tographic algorithms,and selected three deep learning algorithms:BP neural network,convolutional neural network and recurrent neural network.The parameters of network are determined and the corresponding classifiers for identifying cryptographic algorithms are constructed.Identify 8 cryptographic algorithms with the established neural networks.The experimental results show that the three deep learning algorithms can identify the en-cryption algorithms to which the ciphertext belongs with the recognition accuracy rate from 85%to 92%,which has a recognition accuracy rate much higher than that of a ran-dom forest.Finally,analyzed the reason that why the deep learning algorithm is better than the random forest algorithm in the task of identifying cryptographic algorithms.