| The effective recognition of binary cryptographic algorithms can provide important technical support for password cracking and security mechanism analysis,and is of great significance for malicious code detection,information protection,and so on.However,existing recognition technologies have many problems,such as high false alarm rates,low recognition accuracy,poor anti confusion and anti optimization capabilities.Therefore,this article studies the recognition method of binary cryptographic algorithms based on deep learning.The main work and innovative points are as follows:1.Propose a method for static advanced feature extraction of binary encryption algorithms.Based on the order matters method,the disassembly code of the binary encryption function is taken as the input,and the semantic information is extracted from the code Basic block with Bert;Considering the Data dependency of binary encryption algorithm,MPNN is used to extract the global structure information of the graph from the code flow diagram containing data flow and control flow,and CNN is used to extract the local node order information of the graph,so as to improve the accuracy of binary encryption algorithm identification.2.A feature representation method of binary encryption function is proposed,which combines the static advanced features of binary encryption function with the features of Grayscale.In addition to the extraction of static advanced features,the introduction of Grayscale features that can to a certain extent represent the dynamic execution of the binary encryption algorithm improves the anti confusion and anti optimization capabilities of the recognition model of the binary encryption algorithm.3.Propose a two-level recognition method for binary encryption algorithms based on encryption principles and algorithm partitioning.According to the different encryption principles,the first level is roughly divided into four categories:Block cipher,public key cipher,Stream cipher and Hash function,plus non cryptographic function,it is divided into five categories.The second level is subdivided into specific encryption algorithms based on the characteristics of the encryption algorithm,making the recognition of binary encryption algorithms more targeted and the classification results more accurate. |
