Sentinel-1 Enhanced Resolution Image by Deep Learning with TerraSAR-X

Abstract

To improve the quality of SAR images, we proposed to train a deep neural network with TerraSAR-X. This is done by using a Dialectical Generative Adversarial Network (Dialectical GAN) to generate high-quality SAR images. This method is based on the analysis of hierarchical SAR information and the “Dialectical” structure of GAN frameworks. As a demonstration, a typical example will be shown where a low-resolution SAR image (e.g., a Sentinel-1 image) with large ground coverage is translated into a high-resolution SAR image (e.g., a TerraSAR-X image). Three traditional algorithms are compared and a new algorithm is proposed based on a network framework by combining conditional WGAN-GP (Wasserstein Generative Adversarial Network - Gradient Penalty) loss functions and spatial gamma matrices under the rule of dialectics. Experimental results show that the SAR image translation works very well when we visually compare the results of our proposed method with the selected traditional methods. Translation of Sentinel-1 data to TerraSAR-X image resolution has attracted great interest within the remote sensing community. First, the high resolution of TerraSAR-X generates SAR images rich in information that allow new innovative applications. Second, the wide area coverage of Sentinel-1 images reduces the need for multiple acquisitions, and decreases the demand for high-cost data. Third, it is much easier for researchers to access Sentinel-1 images than TerraSAR-X images because the Sentinel-1 images are freely available, while the TerraSAR-X images are usually commercial. For validation, we used images of urban areas, so we can apply a spatial matrix to extract geometrical arrangement information. Our method learns an adaptive loss function based on the image pairs at hand, and is regularized by the prescribed image style, which makes it applicable to the task of SAR image translation. Based on the advantages of using a GAN, we have achieved very good results with detailed visual effects demonstrating that our method is better than the existing traditional methods being compared in our presentation.