Cross - Architecture Binary Code Similarity Detection Method Based on Graph Neural Network

In the IoT realm, the burgeoning number of devices makes firmware security paramount. As software demands evolve, developers reuse code, which may contain flaws. Firmware's closed - source nature and high repair costs exacerbate security issues. A common fix is using defective code for similarity - based scans. However, device diversity in architecture, compiler, etc., complicates this. Binary code similarity analysis based on character and logic features has limitations for cross - architecture detection. This paper thus presents GNOPC, a graph - neural - network - based approach. It combines control flow graphs and operation codes. After obtaining these from binary code, GNN embeds the control flow graph into vectors for similarity judgment. Tests on 7 open - source software with 24 libraries show GNOPC outperforms the latest BCSD method in AUC, recall, and MRR.