Real Fake Image Classification using Explainable EfficientNetV2S: A Comparative Analysis
Vikas Khullar, Rakesh Ahuja, Vikas Solanki, Amit Chaudhary · 2024
This paper investigates the Explainable EfficientNetV2S model's ability to distinguish between authentic and counterfeit photos in order to social protection. By carefully organizing the data, as well as by creatively using explainable images, study analyze the activation layer-wise results of EfficientNetV2S on actual and false image datasets. This study use both traditional CNN and transfer learning technology with a dataset split 80% for training and 20% for testing, with a particular emphasis on the robust EfficientNetV2S pretrained on ImageNet. Metrics like accuracy, precision, recall, and categorical losses are all included in this study, which shows that EfficientNetV2S significantly improves performance. It outperforms the conventional CNN model with remarkable accuracy of 98.99% and validation accuracy of 97.81%. This study’s result shows that the EfficientNetV2S discriminate the fictitious and original images in efficient and effective manner and resulting in advancements in authenticity verification and image classification.