Unsupervised Remote Sensing Image Super-Resolution Method Based on Adaptive Domain Distance Measurement Network
Yangshuan Hou, Jishuai Zhang · 2020 3rd International Conference on Advanced Electronic Materials, Computers and Software Engineering (AEMCSE) · 2020
Compared with supervised learning, unsupervised learning is more practical; however, the associated training process is more difficult and complex. To solve the problems of unstable training and insufficient diversity of generative adversarial networks (GAN), which are widely used to realize unsupervised learning, we propose a novel unsupervised remote sensing image super-resolution method based on a reverse generating network module and the adaptive domain distance measurement network. The discriminant network of GAN is considered as a tool to measure a certain image attribute instead of the original GAN binary classification network. Furthermore, the adaptive domain distance measurement network is used to back feed the information of a high-resolution image to guide the optimization of the generating network. The results of experiments performed on various datasets demonstrate the effectiveness of the proposed method.