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
摘要待审
何静飞 / 北京理工大学
陈亮 / 北京理工大学
师皓 / 北京理工大学
陈宇航 / 北京理工大学
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.

 
重要日期
  • 会议日期

    09月20日

    2024

    09月22日

    2024

  • 08月30日 2024

    初稿截稿日期

  • 09月22日 2024

    注册截止日期

主办单位
山东省人民政府
中国电子学会
承办单位
中国科学院学部
中国科学院空天信创新研究所息
复旦大学
联系方式
移动端
在手机上打开
小程序
打开微信小程序
客服
扫码或点此咨询
Baidu
map