Superresolution Land Cover Mapping Using a Generative Adversarial Network

Superresolution mapping (SRM) is a commonly used method to cope with the problem of mixed pixels when predicting spatial distribution within low-resolution pixels. Central popular SRM pattern model, which utilized represent land cover The use an inappropriate model limits such analyses. Alternative approaches, as deep-learning-based algorithms, learn from training data through convolutional neural network, have been shown considerable potential. Deep learning methods, however, are limited by issues way fraction images utilized. Here, novel based on generative adversarial network (GAN), GAN-SRM, proposed that uses end-to-end address main limitations existing methods. potential GAN-SRM was assessed using four subsets and compared hard classification several experimental results show set methods explored, able generate most accurate high-resolution maps.