Blockchain‐based object detection scheme using federated learning

Abstract

The rapid development of computing devices and automation in various fields drastically increased the growth of data, which promotes the usage of machine learning (ML) techniques to get insights from the generated data. However, data processed by ML algorithms lead to several privacy issues, including leakage of users' biometric data while sharing it through the network to train the object detection model. Therefore, federated learning (FL) was introduced, in which the models are trained locally; only model parameters are shared between central authority (CA) and end nodes. They will eventually maintain a common model for all the participating devices. However, many problems are associated with FL, such as the difference in data consumption rate, training capabilities, geographical challenges, and storage capacity. These problems might lead to differences in the common global model and thus an inefficient FL approach. Moreover, the presence of a CA results in a single point of failure and is vulnerable to various attacks. Motivated by the aforementioned discussion, in this article, we propose a blockchain‐based object detection scheme using FL that eliminates the CA by using distributed InterPlanetary File System (IPFS). Global models can be aggregated periodically when several local model parameters are uploaded on the IPFS. Nodes can fetch the global model from the IPFS. The global aggregated object detection model has been evaluated for various scenarios such as human face detection, animal detection, unsafe content detection, noteworthy vehicle detection, and performance evaluation parameters such as accuracy, precision, recall, and end‐to‐end latency. Compared to traditional models, the proposed model achieved an average accuracy of 92.75% on the object detection scenarios mentioned above.