漏洞检测方法、装置、电子设备、存储介质及程序产品
By preprocessing the binary code, extracting function features using natural language processing and graph embedding neural networks, and combining feature fusion with multilayer perceptrons, the problem of high false alarm rate in existing binary code vulnerability detection methods is solved, achieving higher detection accuracy and efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING UNIV OF POSTS & TELECOMM
- Filing Date
- 2023-04-24
- Publication Date
- 2026-07-17
AI Technical Summary
Existing deep learning-based binary code vulnerability detection methods suffer from high false positive rates, mainly due to biases introduced by human-selected features and the neglect of instruction-level and control flow graph structure information.
By acquiring and preprocessing the code fragments to be detected, the instruction sequence and control flow graph are determined. Function sequence features are extracted using a natural language processing model, and function graph structure features are extracted using a graph embedding neural network. Feature fusion is then performed using a multilayer perceptron, and the similarity of adjacent function feature vectors is calculated for vulnerability detection.
It effectively eliminates the bias introduced by human selection of features, focuses on easily overlooked features, and improves the accuracy and efficiency of vulnerability detection.
Smart Images
Figure CN116663008B_ABST