Research And Application Of Vulnerability Detection Method For Smart Contracts Based On Multi-model Learning Patterns

With the rapid development of computer technology,blockchain,as a new decentralized technology,has begun to be widely used.Currently,blockchain has entered the decentralized application stage,with a large number of decentralized applications(DApps)and new applications such as decentralized finance(DeFi)beginning to develop.The foundation of all decentralized applications is smart contracts.Smart contracts are digital contracts that can be automatically executed and deployed on the blockchain.The blockchain provides a decentralized platform for smart contracts,which use the blockchain as a distributed ledger to record the state and execution results of smart contracts.Due to the automatic execution and decentralized nature of smart contracts,they have been applied in various fields,such as digital currency,digital identity authentication,digital asset management,and decentralized market transactions.However,once smart contracts are deployed on the blockchain,they cannot be changed,and since smart contracts involve digital assets,it is necessary to ensure the correctness and security of the smart contract before deployment.With the rapid development of smart contract technology,the number of smart contracts has exploded,and the security issues of smart contracts have received widespread attention due to economic losses caused by contract vulnerabilities.However,existing smart contract vulnerability detection tools rely too heavily on rigid rules defined by experts,leading to high false positive rates and low accuracy during the detection process.Additionally,the existing vulnerability detection tools also suffer from issues such as cumbersome operation and difficult deployment.Deep learning,as a branch of machine learning,has good generalization ability and adaptability,and can learn common patterns and features in input data.Therefore,based on deep learning and by combining the advantages of different network learning modes,this paper proposes the new smart contract vulnerability detection model.The work completed in this paper is as follows:Firstly,to address the issue of high false positive rates in smart contract vulnerability detection,using the smart contract reentrancy vulnerability as an example,this paper proposes a smart contract vulnerability detection model called SCVSN based on Siamese network.During the model training,SCVSN uses the two sub-networks of the Siamese network to learn the different locations of vulnerabilities in the smart contract,thereby reducing the false positive rate in the detection process.Furthermore,the two sub-networks share parameters in the feature extraction process,which improves the detection efficiency of smart contract vulnerabilities.In addition,to enable the Siamese network to be used for vulnerability detection,this paper improves the training method of the Siamese network and designs an algorithm for constructing positive and negative sample pairs.Secondly,to address the issue of low accuracy in smart contract vulnerability detection,this paper designs a smart contract vulnerability detection model called CBGRU based on a hybrid network.This model combines the advantages of different word embedding models and different deep learning networks for smart contract vulnerability detection.To verify the proposed performance,this paper conducts a series of experiments and compares it with previous research to demonstrate that CBGRU has excellent performance in smart contract vulnerability detection tasks.Thirdly,in response to the issue of complex operation and difficult deployment of existing smart contract vulnerability detection tools,this paper designs a smart contract vulnerability detection system that combines the advantages of the CBGRU and SCVSN models,allowing users to quickly perform smart contract vulnerability detection.Additionally,this paper uses a cross-platform high-level computer language for system design and provides users with an interactive interface for operation.In summary,this paper investigates the smart contract vulnerability detection approach based on deep learning.Through the two proposed intelligent contract vulnerability detection models in the paper,the efficiency of vulnerability detection has been improved,the false positive rate in vulnerability detection has been reduced,the accuracy of vulnerability detection in vulnerability detection has been improved,and an easy-to-use intelligent contract vulnerability detection platform has been designed to enhance user experience.