Comparing the Effectiveness of GAN-based Methods for Enhancing EuroSat Satellite Imagery Classification

Addressing the challenge of limited data in remote sensing, this research investigates the application of Generative Adversarial Networks (GANs) to augment the EuroSat dataset. Four GAN variants (Info-GAN, C-GAN, CDC-GAN, and AC- GAN) were employed to expand a reduced EuroSat dataset by 50%. The augmented datasets were subsequently used to train Convolutional Neural Networks (CNNs) for image classification. Comprehensive evaluation metrics (accuracy, recall, precision, Fl-score, AUC, IS, and FID) were employed to assess the performance of the generated images and the trained models. Results indicate that CDC-GAN exhibited the highest classification accuracy and MCC, while AC-GAN excelled in ROC-AUC, precision, and recall. This study underscores the potential of GAN s as a valuable tool for enhancing remote sensing image datasets and improving classification performance.