A Federated Attention-Based Multimodal Biometric Recognition Approach in IoT.

Area of Science:

• Computer Science
• Artificial Intelligence
• Cybersecurity

Background:

• The proliferation of Internet of Things (IoT) devices necessitates robust access control mechanisms.
• Unimodal biometric recognition systems face limitations in accuracy and robustness for IoT applications.
• Data privacy concerns and regulations hinder the collection of training data for IoT security models.

Purpose of the Study:

• To develop an advanced multimodal biometric recognition system for secure IoT access control.
• To enhance feature extraction and fusion using attention mechanisms in biometric recognition.
• To address data privacy challenges in IoT by employing Federated Learning for model training.

Main Methods:

• Proposed an attention-based multimodal biometric recognition (AMBR) network integrating attention mechanisms.
• Utilized Federated Learning (FL) to train the AMBR model collaboratively across distributed data sources.
• Evaluated the AMBR network's performance on the VoxCeleb1 dataset across three trial lists.

Main Results:

• Achieved low Equal Error Rates (EER) of 0.68%, 0.47%, and 0.80% on the VoxCeleb1 trial lists.
• Demonstrated superior performance compared to existing state-of-the-art biometric recognition methods.
• Experimental results in Federated Learning settings confirmed the efficacy of the AMBR-FL approach.

Conclusions:

• The proposed AMBR network effectively extracts and fuses multimodal biometric features for improved recognition.
• Federated Learning enables privacy-preserving training of biometric recognition models for IoT systems.
• The AMBR network combined with FL presents a promising solution for secure and privacy-conscious IoT access control.