Multi-Modality Translation from SAR Image to Optical Image for Remote Sensing Applications
Hao Zhou, Xiao Xiao, Yilong Lu · 2021
Synthetic Aperture Radar (SAR) is an indispensable remote sensing technology nowadays. However, compared to conventional optical imageries, SAR images have significant drawbacks in human interpretation, especially for those without expert knowledge. In this paper, we explore a new method to enhance the human interpretation for raw SAR imageries by turning SAR data into readable optical imageries through a multi-modality translation procedure based on a deep neural network that consists of a novel two-step architecture. To illustrate the concept, remote sensing data from Sentinel-1 and Sentinel-2 is used for training and validation. The results show significant improvement in ease of interpretation from the original SAR images with clear views of different land covers such as fields, water, highways, and etc. The statistical performance of SSIM and PSNR has achieved 0.42 and 19.10 dB respectively, indicating a promising translation quality.