Research on Vulnerability Detection Techniques Based on Static Analysis and Program Slice

Research on Vulnerability Detection Techniques Based on Static Analysis and Program Slice

연구 분야: Strategies

학회: 2024 6th International Conference on Electronic Engineering and Informatics (EEI)

Binary program vulnerabilities are particularly difficult to detect and understand using traditional methods due to their concealment and complexity. To address this challenge, we propose an innovative binary code vulnerability detection frame- work that combines program slicing techniques with a hybrid neural network model to achieve high-precision identification and classification of vulnerabilities in binary code. Specifically, we first generate semantically relevant binary slices using Program Dependence Graphs (PDG). These slices are then processed by a Bidirectional Long Short-Term Memory Network (BLSTM) and an attention mechanism to extract deep semantic features, while a Convolutional Neural Network (CNN) is employed to extract sequential features of the slices. Additionally, we introduce a Siamese network to evaluate the similarity between different code fragments, enhancing the model's ability to recognize unknown vulnerability patterns. Experimental results demonstrate that our approach performs exceptionally well in binary program vulnerability detection, significantly improving both the accuracy and efficiency of the detection process.