Deep Learning-based Art Authentication: A CNNModel with Grad-CAM for AI-Generated Image Detection

Objectives: To develop a CNN-based system for detecting AI-generated images in art, aimed at enhancing art authentication. The study proposes an automated and scalable approach to identify forgeries effectively, reducing reliance on manual inspection. Methods/Statistical analysis: The proposed system integrates Convolutional Neural Networks (CNNs) with Generative Adversarial Networks (GANs) and ReLU activation functions. GANs generate synthetic forgery datasets, enhancing CNN's capacity to identify nuanced features of forgeries. Extensive preprocessing, including resizing, augmentation, and dropout layers, ensures improved generalization. The system’s performance was tested on a dataset of 23,000 images, achieving significant classification accuracy with minimal overfitting. Findings: Experimental results demonstrate an accuracy of 82.47%, with the system effectively distinguishing AI-generated and human-created artworks. The addition of GANs improved dataset diversity, while ReLU activation enhanced model learning and reduced computational costs. Validation metrics confirmed robust model generalization, suggesting suitability for realworld deployment in art verification scenarios. Novelty/Applications: By leveraging GANgenerated training datasets and incorporating ReLU, this model provides a reliable, automated solution for art authentication, suitable for experts, collectors, and institutions. Its scalability and efficiency position it as a critical tool for preserving cultural heritage and combating forgery.