Related Experiment Video
Updated: May 28, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
455
FL-TENB4: A Federated-Learning-Enhanced Tiny EfficientNetB4-Lite Approach for Deepfake Detection in CCTV Environments
Jimin Ha1, Abir El Azzaoui1, Jong Hyuk Park1
1Department of Computer Science and Engineering, Seoul National University of Science and Technology, Seoul 01811, Republic of Korea.
Sensors (Basel, Switzerland)
|February 13, 2025
Summary
This study introduces FL-TENB4, a new system for detecting deepfakes in CCTV footage using Tiny Machine Learning (TinyML) and Federated Learning (FL). It offers real-time, privacy-preserving deepfake detection for resource-limited cameras.
Area of Science:
- Computer Vision and Artificial Intelligence
- Cybersecurity and Privacy
Background:
- CCTV systems are crucial for public safety but vulnerable to deepfake manipulation.
- Existing deepfake detection methods are computationally intensive, hindering real-time CCTV application.
- Deepfakes threaten the integrity of video evidence and personal privacy.
Purpose of the Study:
- To develop an efficient and privacy-preserving deepfake detection framework for CCTV.
- To address the limitations of current deepfake detection solutions in resource-constrained environments.
Main Methods:
- Proposed FL-TENB4 framework integrating Tiny Machine Learning (TinyML) with EfficientNetB4-Lite.
- Utilized Federated Learning (FL) for privacy-preserving, collaborative model training.
- Implemented a lightweight model optimized for edge devices and real-time processing.
Main Results:
- FL-TENB4 demonstrated high deepfake detection accuracy on the FaceForensics++ dataset.
- Achieved significantly reduced model size and low inference latency.
- Validated suitability for real-world, resource-constrained CCTV environments.
Conclusions:
- FL-TENB4 offers an effective solution for real-time deepfake detection in CCTV systems.
- The framework balances performance, efficiency, and data privacy.
- Enables enhanced security and reliability of surveillance systems against deepfake threats.
