An Efficient Deepfake Detection System Using ConvoReinAutoNet and GeoFisherNet

This study suggests a hybrid optimization model and a new deep learning technique to create an effective deepfake detection system. During the preprocessing stage, deepfake database images are improved by employing Gaussian Filter and Histogram Equalization and are ready for analysis. The recently proposed Improved Local Ternary Patterns (I-LTP) approach collects textural and temporal information for feature extraction. The advanced GeoFisherNet, which effectively integrates spatial and temporal properties, is then utilized to fuse these data. The Marine Predator Customized White Shark Optimizer (MCWO), a hybrid approach that combines the White Shark Optimization Algorithm (WSO) and Marine Predator Algorithm (MPA), is used to find the most discriminative features during the feature selection phase. Lastly, ConvoReinAutoNet (CRAN), a revolutionary deep learning architecture that combines Convolutional Neural Networks (CNN), Deep Reinforcement Learning (DRL), and Autoencoders, is applied to the fused and optimized features in the classification phase to make precise detection decisions. The Python implementation of the suggested system shows better detection efficiency and accuracy of two data splits (70% and 80%) are 98.78% and 99.42% than the current methods.