Deepfake detection method based on reinforcement learning DQN algorithm

Abstract

The invention relates to the technical field of machine learning, and concretely relates to a Deepfake detection method based on a reinforcement learning DQN algorithm. The method comprises the following steps: 1, collecting sample data, and dividing the sample data into a training set S and a test set T; 2, inputting the training set S into the Q network, inputting [state-action pair] (si, ai) of the training set S and Q (si, a'i) output by the Q network into a discriminator D to obtain confidence delta, deriving and updating a model parameter theta i of the Q network by using the confidence delta, and obtaining a Q network detection model; 3, testing a Q network; and 4, applying the Q network detection model to the true and false discrimination of Deepfake. The Q network is trained by using a group of true and false known samples through the reinforcement learning DQN algorithm, the Q value is updated through the reinforcement learning DQN algorithm, finally, the Q network is trained into a model capable of judging the true and false of a video or a picture, a complex frame structure does not need to be designed, the generalization ability is high, and the application scene is wide.

Description

technical field [0001] The invention relates to the technical field of machine learning, in particular to a deepfake detection method based on reinforcement learning DQN algorithm. Background technique [0002] Deepfake is a combination of deep learning (Deep-learning) and fake (fake). It can superimpose the image of the target person's face on the corresponding position of the original person's face in the video, thereby creating a video containing the target person. With the continuous strengthening of Deepfake technology, It is already difficult for people to distinguish some fake pictures or videos with the naked eye, and there are a large number of deepfake videos on the Internet. The Dutch network security company DeepTrace found more than 14,000 deepfake videos in 2019, an increase of 84% over 2018. %, the technology is increasingly being misused. Therefore, it is particularly important to detect these fake pictures or videos. [0003] The current deepfake detection...

AI Technical Summary

Problems solved by technology

[0006] In order to overcome the problems that the existing deepfake detection technology model training process is complex and easy to overfit, and the application scenarios are not comprehensive enough, the present invention provides a deepfake detection method based on reinforcement learning DQN algorithm

Images