A WGAN-GP Framework for SAR and Optical Remote Sensing Image Fusion

Synthetic Aperture Radar (SAR) and optical images are widely used in remote sensing applications due to their unique capabilities and complementary data. SAR can penetrate through cloud cover and adverse weather conditions, while optical images provide rich spectral information, however are affected by cloud cover. This paper proposes a novel approach for SAR and optical image fusion using the Wasserstein Generative Adversarial Network with gradient penalty (WGAN-GP). In addition, we introduce the content loss function that combines the structural similarity loss in the literature, with the pixel loss and gradient loss for optimal fusion. Experiments on the Sentinel dataset show that the proposed methodology outperforms existing state-of-art architectures based on CNN and wavelet transform. We also illustrate, how the fused image is more beneficial for both human perception and automatic computer analysis.