126 / 2024-08-20 15:48:23

A Dual Network Approach with Enhanced Feature Guidance for SAR-to-Optical Image Translation

Synthetic aperture radar,generative adversarial networks,SAR-to-image translation

Synthetic Aperture Radar (SAR) images own dominant merits, such as all-weather and all-day working conditions,but it is difficult for unprofessional people to interpret SAR images. Translating SAR images into optical images to assist interpretation can facilitate the transformation of SAR data into usable information. Therefore, an advanced SAR to Optical (S2O) image translation method utilizing a parallel Generative Adversarial Network (GAN) framework was proposed. This method incorporates an optical image reconstruction network alongside the S2O translation network, enhancing the fidelity and color consistency of the translated images. By employing a domain alignment module and a deep supervision feature loss module, the network effectively utilizes abundant optical image features to overcome the scarcity of SAR-optical image pairs. Experiments conducted on the SEN1-2 dataset demonstrate superior performance over existing methods, particularly in preserving detailed structural information in images.