Deepfake Video Detection Based on EfficientNet-V2 Network.

Area of Science:

• Computer Science
• Artificial Intelligence
• Cybersecurity

Background:

• Deep learning advancements enable sophisticated fake media creation.
• Malicious use of deepfakes (e.g., political disinformation, financial fraud) poses societal risks.
• Existing fake face detection systems require enhanced performance.

Purpose of the Study:

• To develop an advanced deepfake detection system.
• To leverage EfficientNet-V2 for improved authenticity verification of images and videos.
• To mitigate the societal impact of sophisticated fake media.

Main Methods:

• Implementation of the EfficientNet-V2 network for deepfake detection.
• Training and testing the model on two large-scale, mainstream fake face datasets.
• Comparative analysis against existing deepfake detection networks.

Main Results:

• EfficientNet-V2 demonstrated superior performance compared to existing detection networks.
• The proposed method achieved high accuracy in distinguishing real from fake faces.
• Successful detection of real-world images and videos with excellent visualization.

Conclusions:

• EfficientNet-V2 is a highly effective architecture for deepfake detection.
• The enhanced detection system contributes to combating the spread of misinformation.
• The study provides a robust solution for verifying media authenticity.