Research On Smart Contract Vulnerability Detection Method Based On Deep Learning Technology

Blockchain technology is a decentralized technology that allows users to conduct trusted asset transactions with multiple parties without the need for third-party intervention.As a result,blockchain technology has been widely promoted and applied.Smart contracts are snippets of code deployed in a distributed independent ledger that automates traditional contract code through computer instructions.Smart contracts have received widespread attention due to their decentralization and tamper-proof features.Users can establish contractual relationships with their partners through the execution of smart contracts,while promoting the fulfillment and maintenance of the contract.Smart contracts are therefore widely used in fields such as finance,healthcare,and new energy.In recent years,Ethereum has experienced frequent security incidents such as asset theft,which is mainly due to the explosive growth in the number of smart contracts,leading to a multitude of different types of security vulnerabilities.Hackers can easily exploit these vulnerabilities to generate unlimited virtual currency and transfer it at will.The frequent occurrence of security incidents not only disrupts the order of the capital market but also raises questions about the security status of Ethereum among users,ultimately affecting the further development of blockchain technology.The types of existing smart contract vulnerabilities are becoming increasingly complex and unpredictable.The most discussed known types of smart contract vulnerabilities in current research include integer overflow,reentrancy,infinite loop,timestamp dependency,callstack depth attack,etc.The research in this article is also focused on these vulnerability types.The widespread application of smart contracts in various fields provides increasingly available training data for existing research.Designing a highly generalizable and effective smart contract vulnerability detection method is currently a research focus for smart contract security issues,and is also the purpose of this research.Therefore,this article deeply explores the method of smart contract security detection by combining the characteristics of deep learning technology and ensemble learning technology in the context of blockchain.The main work includes:Firstly,a proposed intelligent contract vulnerability detection method combines deep learning technology based on neural network architecture with a Support Vector Machine(SVM)classifier.By processing and analyzing the source code of smart contracts,a smart contract vulnerability detection model with high classification accuracy and good generalization is designed.Some existing studies have overlooked the advantage that SVMs have a solid mathematical theoretical foundation.Using them as classifiers can not only effectively solve the problem of building high-dimensional data models with limited samples,but also overcome the shortcomings of neural network-based classification in terms of not obtaining global optimal solutions and poor interpretability.The study presents the specific structure and detection process of the smart contract vulnerability detection method with a focus on reentrancy vulnerabilities.A significant amount of experimental data shows that the feature data extracted from deep learning models based on neural network architecture can be processed and classified by SVM classification algorithm to further improve the generalization ability and vulnerability detection accuracy of vulnerability detection models.It is demonstrated that the vulnerability detection capability of the model can be improved to a certain extent by using a classifier with high classification power to process the feature data based on the effective learning of vulnerability features,which provides a research basis for the smart contract vulnerability detection method based on deep learning technology in this paper.Secondly,this research proposes a smart contract vulnerability detection method based on the SPCBIG-EC model,which combines the serial-parallel extraction structure and ensemble classification.The method includes a multi-level analysis of vulnerability features using a semantic information extraction technique based on the serial-parallel structure and a multi-dimensional analysis of vulnerability classification using an ensemble classification algorithm based on smart contract feature parameters.In this approach,a cascade method based on the combination of structured spatial convolutional network and serialized temporal recurrent network is first designed to achieve multi-feature combination of temporal structure and position information of key features.Secondly,a multi-scale serial-parallel convolution structure is proposed to address the problem of information loss caused by pooling operations in convolutional networks.The word vector representation of smart contracts is enriched and features are fused under multi-scale convolution to ensure the integrity of key information.Finally,to address the issue of high misclassification rates when the dataset has a small number of samples,a multi-angle analysis of intelligent contract feature parameters and vulnerability classification technology based on ensemble learning algorithms is proposed.This method updates the data weights by iterative learning to obtain multiple optimized prediction functions.It makes the bounds of the classification model more stable and improves the robustness of the classifier.Thirdly,a smart contract vulnerability detection system called D&E is proposed.The system is capable of automatically detecting potential threat vulnerabilities in Ethereum smart contract code.After logging into the testing end of the D&E intelligent contract security vulnerability detection system and determining the feature extraction and classification methods,the intelligent contract code is uploaded.Then,a test request for intelligent contract vulnerability testing is sent to the test object,and the security detection tool intercepts the data of the test request.Based on the vulnerability detection strategy,the detection tool performs the detection and obtains the detection results of the intelligent contract vulnerability.The detection results include information on whether a vulnerability exists and the type of vulnerability.This article focuses on researching the security issues of Ethereum smart contracts.With the current rapid development of deep learning techniques and ensemble learning methods,we try to apply them to the vulnerability detection of smart contract source code to achieve a new data-driven automated intelligent detection task.In summary,this paper investigates a deep learning approach to smart contract vulnerability detection based on a neural network architecture.The two proposed methods and detection system in this paper improved the detection efficiency,generalization,and robustness of the vulnerability detection model.This has significant significance for the extension of the smart contract vulnerability detection task in new fields.