Recently, artificial intelligence, deep learning and Generative Adversarial Networks (GANs) adaptabilities for deepfake detection and forensics have become an emerging field of research interest. GANs have been widely studied since it was first proposed, and many applications have been produced to generate contents such as videos and images. The application of these new technologies in many fields makes it more and more difficult to distinguish between true and fake content. This study analyzes more than hundred published papers related to the application of GANs technology in various fields to generate digital multimedia data and expounds the technologies that can be used to identify deepfakes, the benefits and threats of deepfake technology, and how to crack down deepfakes. The findings indicate that although deepfakes pose a major threat to our society, politics and commerce, a variety of means are listed to limit the production of unethical and illegal deepfakes. Finally, the study also puts forward its limitations and possible future research directions and recommendations.
Different CNNs models do not perform well in deepfake detection in cross datasets. This paper proposes a deepfake detection model called DeepfakeNet, which consists of 20 network layers. It refers to the stacking idea of ResNet and the split-transform-merge idea of Inception to design the network block structure, That is, the block structure of ResNeXt. The study uses some data of FaceForensics++, Kaggle and TIMIT datasets, and data enhancement technology is used to expand the datasets for training and testing models. The experimental results show that, compared with the current mainstream models including VGG19, ResNet101, ResNeXt50, XceptionNet and GoogleNet, in the same dataset and preset parameters, the proposed detection model not only has higher accuracy and lower error rate in cross dataset detection, but also has a significant improvement in performance.
The purpose is to solve the problems of large positioning errors, low recognition speed, and low object recognition accuracy in industrial robot detection in a 5G environment. The convolutional neural network (CNN) model in the deep learning (DL) algorithm is adopted for image convolution, pooling, and target classification, optimizing the industrial robot visual recognition system in the improved method. With the bottled objects as the targets, the improved Fast-RCNN target detection model's algorithm is verified; with the small-size bottled objects in a complex environment as the targets, the improved VGG-16 classification network on the Hyper-Column scheme is verified. Finally, the algorithm constructed by the simulation analysis is compared with other advanced CNN algorithms. The results show that both the Fast RCN algorithm and the improved VGG-16 classification network based on the Hyper-Column scheme can position and recognize the targets with a recognition accuracy rate of 82.34%, significantly better than other advanced neural network algorithms. Therefore, the improved VGG-16 classification network based on the Hyper-Column scheme has good accuracy and effectiveness for target recognition and positioning, providing an experimental reference for industrial robots' application and development.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.