IoT‐Integrated Detection and Classification of Deepfake Images and Videos Using Custom Deep Learning Models

Deep fakes are becoming more common; they include editing previously published films and photos to produce content that appears authentic but is wholly fake. The development process has been considerably expedited by the widespread availability of deep learning techniques, such as autoencoders, generative adversarial networks, and user-friendly software. These sophisticated algorithms adeptly fuse and modify visual and audio elements, facilitating the production of content that closely mimics genuine footage, even for those without specialized knowledge. The malicious manipulation of images and videos poses significant security and societal concerns. With an emphasis on facial alteration, the goal of this research is to create a deep learning system integrated with Internet of Things (IoT) for the detection and classification of deepfake images and videos. The dataset used for the project is either FaceForensics++, Celeb-DF, or the Deepfake Detection Challenge Dataset, available on Kaggle, consisting of real and deepfake images and videos. By utilizing recurrent and convolutional neural networks (CNNs), we have made developments in deepfake detection. Commencing with preprocessing the data, extracting frames from the videos, and separating the dataset into training and validation sets, IoT devices continuously capture and transmit data for real-time analysis. For the detection and classification of deepfake images and videos, OpenCV and face recognition are used for facial detection, whereas CNNs are used by the system to extract features at the frame level. A recurrent neural network is trained using these features. Various techniques such as data augmentation, learning rate scheduling, and early stopping enhance model performance. This comprehensive IoT-integrated approach ensures accurate discrimination between authentic and deepfake content, addressing concerns regarding the integrity of digital media.