NFT仿冒欺诈的测量与检测技术  

Measurement and Detection Techniques for NFT Counterfeiting Fraud

作　　者：廖鹏[1] 方滨兴[1,2] 刘潮歌 王志 张云涛 崔翔 LIAO Peng;FANG Bin-Xing;LIU Chao-Ge;WANG Zhi;ZHANG Yun-Tao;CUI Xiang(Key Laboratory of Trustworthy Distributed Computing and Service(Beijing University of Posts and Telecommunications),Ministry of Education,Beijing 100876;School of Cyber Science and Technology,Guangzhou University,Guangzhou 510006;Zhongguancun Laboratory,Beijing 10094)

机构地区：[1]北京邮电大学可信分布式计算与服务教育部重点实验室,北京100876 [2]广州大学网络空间安全学院,广州510006 [3]中关村实验室,北京100094

出　　处：《计算机学报》2024年第5期1065-1081,共17页Chinese Journal of Computers

摘　　要：近年来非同质化代币(Non-Fungible Token,NFT)繁荣发展,但安全问题也日益凸显,尤其是NFT的仿冒问题.在去中心化的环境下,仿冒已有的NFT作品变得相对容易,而辨别真伪却尤其困难.本文围绕仿冒NFT的测量与仿冒检测方法的评估进行了系统深入的研究.建立了包括形式化定义、仿冒过程和仿冒特征在内的NFT仿冒威胁模型,给出了 NFT仿冒定义,分析了 NFT仿冒方式,给出了判定仿冒NFT的一般性方法.大规模采集了全球最大的NFT交易平台OpenSea上50 000个NFT项目的智能合约地址和历史交易数据,并从以太坊区块链上采集了这些NFT项目的名称、创建时间、元数据以及链下存储的NFT图像数字载体,从中选取668个交易量排名靠前的NFT项目围绕NFT仿冒问题开展了测量工作,结果表明其中95个项目被仿冒248次,交易金额超过2600万美元,足见NFT生态所面临的仿冒欺诈问题之严重.本文采用了 22种图像数据增强方法,构造了 5000个扰动较小的攻击测试样本数据集,评估了 OpenSea和知名的第三方商业检测平台Fnftf对仿冒NFT检测的鲁棒性,测试结果表明有6种图像数据增强方法构造的攻击测试样本能够轻易绕过检测,揭示了 NFT行业仿冒欺诈检测产品的脆弱性.为提高对仿冒NFT检测的鲁棒性,本文提出并实现了一种基于深度学习的NFT图像仿冒检测模型,实验表明其AUC值相较于Fnftf提升了 15.9%.With the boom in non-fungible tokens(NFT),security issues are becoming increas-ingly prominent,one of which is the phenomenon of NFT counterfeiting.In the decentralized environment,it becomes relatively easy to counterfeit other creators'NFT artworks,while it is very difficult to identify the authenticity,and identifying fakes and counterfeits requires a high level of blockchain technology background,The most prominent security problem of the current NFT is the threat of legitimacy due to the lack of regulation of NFT projects,and the NFT security problems of infringement,counterfeiting,plagiarism are difficult to solve only by the blockchain's own security mechanisms.In this paper,a systematic and in-depth research is conducted on measuring NFT counterfeiting and evaluating counterfeiting detection methods.A NFT counterfeiting threat model including the formal definition,counterfeiting process and counterfeiting features are established,NFT counterfeiting definition is given,NFT counterfeiting methods are analyzed,and a general NFT counterfeiting detection method is given.The smart contract addresses and historical trading data of 50000 NFT projects on OpenSea,the largest NFT marketplace,were collected on a large scale,and the names,creation times,metadata,and digital payload of NFT images stored off-chain were collected from the Ethereum blockchain,from which 668 top NFT projects in terms of trading volume were selected to conduct measurements around the problem of NFT counterfeiting,and the results showed that 95 of them had been counterfeited 248 times,with the trading volume of more than$26 million,which is a clear indication of the seriousness of the counterfeiting and fraud problem facing the NFT ecosystem.Twenty-two image data augmentation methods were used to construct 5000 less perturbed attack test sample datasets,and the OpenSea online real-time counterfeit fraud detection system and the well-known third-party commercial detection platform Fnftf were tested for robustness through the black-box testing

关 键 词：非同质化代币 区块链 以太坊 深度学习 对抗攻击 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]