基于深度学习的深度伪造检测方法

Deep Forgery Detection Method Based on Deep Learning

目前DeepFake这种深度伪造技术在网络上被滥用,来制作色情电影、虚假新闻,甚至可能将其用于政治人物来制造政治谣言,这对国家安全和社会稳定带来了潜在的威胁。针对这个问题,提出了一种基于深度学习的深度伪造检测方法。该模型是基于深度神经网络模型EfficientNet-B4,并结合了Vision Transformer技术。实验阶段将该方法在两个数据集FaceForensics++和Celeb-DF上进行了论证。实验结果表明,该方法较之前提出的方法不仅明显提高了识别的准确率,达到了96%,同时还具有很好的泛化性能,在跨数据集方面也取得了72%的识别准确率。该方法具有优良的性能。

This paper proposes a deep forgery detection method based on deep learning.The model is based on the deep neural network model EfficientNet-B4 combined with Vision Transformer technology.In the experimental phase,the method is demonstrated on two frequent datasets,FaceForensics ++ and Celeb-DF.The experimental results show that this method improves the recognition accuracy significantly,reaching 96% compared with the previously proposed method.At the same time,it has good generalization performance, and has achieved a recognition accuracy of 72% across datasets.This method has excellent performance.