Binary vulnerability mining technology based on neural network feature fusion

The high complexity of software and the diversity of security vulnerabilities have brought severe challenges to the research of software security vulnerabilities Traditional vulnerability mining methods are inefficient and have problems such as high false positives and high false negatives, which can not meet the growing needs of software security. To solve the above problems, this paper proposes a binary vulnerability mining technology based on neural network feature fusion. Firstly, this method constructs binary vulnerability data sets containing multiple vulnerability types, then decompile them to the pcode intermediate language level, and then extracts relevant feature vectors from binary vulnerability data sets according to Bert fine tuning model and bilstm model respectively. In order to fully obtain the semantic information of vulnerabilities, this method standardized the two, fused them, and carried out relevant experiments. The experimental results show that the accuracy of vulnerability detection on SARD data set is 96.92%, which is higher than other binary vulnerability detection methods based on neural network.