Traditional fingerprint recognition, relying on large datasets and machine learning, often faces accuracy issues due to data heterogeneity and privacy concerns. This study proposes AI-Fed-FR, a novel fingerprint recognition algorithm using explainable AI-based federated learning to enhance accuracy while ensuring privacy. AI-Fed-FR improves global model performance by iteratively aggregating parameters from user devices and employs explainable AI for denoising low-resolution fingerprint images. A storage sampling-based client scheduling technique addresses client imbalance. Experiments on three real-world datasets show AI-Fed-FR achieving 5.32% higher accuracy than local learning and 8.56% higher than average-based federated learning.

• 변해원 | Haewon Byeon Corresponding author