Embedded Operating System Vulnerability Detection Technology Based on Graph Neural Networks

Due to the characteristics of embedded devices, conventional dynamic and static vulnerability detection is not suitable for embedded operating systems. At present, most of the datasets used in public vulnerability detection research are manually generated, and there is a high repetition rate of data, which is difficult to reflect real-world vulnerabilities. This paper analyzed the typical vulnerability characteristics of embedded operating systems, and designed a graph embedding method suitable for code representation. Based on the theoretical basis of graph neural network (GNN), two vulnerability detection models were designed, and trained using parallel datasets from the real world. The experimental results indicate that the models have a good ability to express and understand the vulnerability characteristics of the embedded operating system, which verifies the effectiveness of GNN in the vulnerability detection of the embedded operating system, and realizes an engineering vulnerability detection technology suitable for the embedded operating system, which has strong practical significance.