Discovering Attacks against Smart Contracts Using Opcode Sequences with Feature Fusion

Abstract

Smart contracts have revolutionized blockchain technology by enabling efficient, automated transactions without intermediaries. However, their increasing prevalence has made them attractive targets for malicious actors, highlighting the critical need for robust security measures. While significant efforts have focused on identifying vulnerabilities in smart contract bytecode before deployment, there still needs to be a substantial gap in detecting execution attacks post-deployment. This gap poses a severe risk to the integrity and security of blockchain transactions. To address this challenge, we introduce an innovative multi-dimensional feature fusion technique for detecting attacks during smart contract execution. Our method comprehensively analyzes opcode behavior across four key dimensions: target operation object, system resource utilization, category, and gas consumption. This multi-faceted approach allows for a more thorough and nuanced detection of potential threats. Our approach demonstrates exceptional performance, achieving a 97.21% detection accuracy and a 97.21% weighted F1-Score. These metrics indicate high precision and recall in identifying various potential attacks.