Conventional centralized image classification methods faced challenges due to data privacy issues and limitations in computing resources, rendering them inadequate for practical applications. Existing federated learning frameworks relied on central servers, and there were security challenges such as single points of failure and data poisoning attacks. To address these issues, this paper proposed a novel image classification scheme that combined privacy-preserving federated learning and blockchain technology. The scheme achieved reliability and security in image classification tasks within a distributed environment. This approach trained the image classification model through federated learning and uploaded to the blockchain network for verification and consensus. During the classification phase, the model obtained the final classification result through weighted combination. Experimental results demonstrate that the proposed scheme ensures the accuracy of image classification while protecting user privacy. In conclusion, this paper provides an effective approach to address data privacy and security concerns in image classification, and presents a positive exploration towards improving classification accuracy.

[2023-08-11] Accepted Paper /doc/2023.06.0246/v1

[2024-02-05] Printed Article /doc/2023.06.0246/v2