Deep Learning-Based Blockchain–IPFS Architecture for Enhancing the Scalability and Trustworthiness of Copyright Protection Systems

This paper addresses the fundamental limitation of existing blockchain–IPFS(InterPlanetary File System) copyright protection systems, which rely solely on hash-based verification and therefore fail to detect semantic tampering and re-encoding attacks. To overcome this gap, we experimentally integrate CNN and Autoencoder-based deep learning models into the on-chain/off-chain architecture and develop a functional prototype. The proposed model learns high-dimensional feature embeddings of digital content, enabling precise identification of subtle edits, compression artifacts, and visually preserved manipulations that conventional systems cannot distinguish. Experimental results demonstrate significantly superior detection performance compared to hash-based methods under re-encoding, resolution changes, and partial manipulation scenarios. Consequently, this study presents a distinctive contribution by positioning deep learning as the core verification engine within a next-generation blockchain–IPFS tamper-detection framework.