Exploring generative models for remote sensing: a comprehensive review

Remote sensing enables systematic observation of the Earth’s surface through aerial and satellite imagery, supporting a wide range of applications from environmental monitoring to urban planning. Recent advances in deep learning models have significantly enhanced the analysis of remote sensing data; however, these models often require large volumes of high-quality labelled data, which are difficult and costly to obtain due to limitations in sensor resolution and data acquisition. Generative Adversarial Networks (GANs), a class of deep generative models, have emerged as a transformative solution by synthesizing realistic data and improving model robustness in data-scarce environments. In addition to data augmentation, GANs facilitate critical remote sensing tasks such as image super-resolution, cloud removal, cross-modal translation, and change detection. Their ability to model complex data distributions and perform adversarial training makes them particularly effective for addressing challenges such as noise, resolution gaps, and domain discrepancies. This review provides a comprehensive overview of GAN architectures, explores their diverse applications in remote sensing, discusses relevant evaluation metrics, and highlights key challenges and opportunities for future research.