Detection of Vulnerabilities in Cryptocurrency Smart Contracts Based on Image Processing

Abstract

The rate of use of cryptocurrencies through smart contracts and decentralized applications remains continually increasing. Ethereum is particularly gaining popularity in the blockchain community. In this work, the authors are interested in retraining vulnerability and timestamping. They propose a detection method based on the transformation of contracts into images and the processing of the latter using Simhash and n-gram techniques to obtain our contracts into images of size 32*32. They combine a technique to preserve the useful characteristics of images for exploitation. Training carried out with the convolutional neuronal network (CNN) model on a sample of 50 normal contracts, 50 contracts vulnerable to retraining, and 33 vulnerable to timestamping gave an accuracy of 88.98% on the detection of vulnerable contracts. The singular value decomposition (SVD) technique is capable of efficiently extracting from images, the key features that characterize contracts in Ethereum.