| Summary: | Synthetic Aperture Radar (SAR) is an indispensable remote sensing technology
nowadays. However, due to the different imaging theory applied in Synthetic
Aperture Radars, the interpretation of SAR images may come out as extremely
different from conventional optical satellite images. Thus, to tackle with the
interpretation problem, a specialized SAR-optical image translation model is developed to directly translate the original SAR images into equivalent optical
satellite images. This model is implemented with a novel two-step Generative Adversarial Network architecture. To present the performance of proposed
model on SAR-Optical image translation task, remote sensing data acquired
from Sentinel-1 and Sentinel-2 is utilized for the model training and validation
phase. The final results indicate a promising performance both on enhancing
the human perception of translated optical images and increasing the statistical indices of PNSR and SSIM, which have reached at 19.09 dB and 0.4211
respectively.
|
|---|