Embedded Operating System Memory-Related Vulnerabilities Detection Based on Taint Analysis and Graph Neural Networks

Embedded operating systems are highly customizable, with limited public information and severe security vulnerabilities. Among the vulnerabilities, one-third are memory-related. Traditional detection methods are inefficient and costly for addressing these vulnerabilities. Deep learning-based approaches may cause information loss or introduce irrelevant noise during the code representation process. Moreover, memory-related vulnerability patches are often very similar to the source code, making it difficult to represent vulnerability features. To address these issues, this paper proposes a graph neural network-based detection method using taint analysis. By tracking and slicing the representation graph based on taint information, the method can expand effective graph information and remove irrelevant noise. Experimental results show this method achieves an F1 score of 81.82% on an embedded OS memory-related Vulnerability dataset.