A blockchain-enabled horizontal federated learning system for fuzzy invasion detection in maintaining space security

Abstract

Recent advances in Industry 4.0 technologies drive robotic objects' decentralisation and autonomous intelligence, raising emerging space security concerns, specifically invasion detection. Existing physical detection methods, such as vision-based and radar-based techniques, are ineffective in detecting small-scale objects moving at low speeds. Therefore, it is worth investigating and leveraging the power of artificial intelligence to discover invasion patterns through space data analytics. Additionally, fuzzy modelling is needed for invasion detection to enhance the capability of handling data uncertainty and adaptability to evolving invasion patterns. This study proposes a Blockchain-Enabled Federated Fuzzy Invasion Detection System (BFFIDS) to address these challenges and establish real-time invasion detection capabilities for edge devices in the low earth orbit. The entire model training process is performed over the blockchain and horizontal federated learning scheme, securely reaching consensus in model updates. The system's effectiveness is examined through case analyses on a publicly available dataset. The results indicate that the proposed system can effectively maintain the desired invasion detection performance, with an average Area Under Curve (AUC) value of 0.99 across experimental runs. Utilising the blockchain-based federated learning process, the total size of transmitted data is reduced by 89.5 %, supporting the development of lightweight invasion detection applications. A closed-loop mechanism for continuously updating the space invasion detection model is established to achieve high space security.