A SAR-to-Optical Image Translation Method Based on PIX2PIX.

Zongcheng Zuo,Yuanxiang Li
DOI: https://doi.org/10.1109/igarss47720.2021.9555111
2021-01-01
Abstract:Optical remote sensing images are susceptible to adverse weather effects, such as cloud occlusion, which lead to low availability of optical image data. However, synthetic aperture radar (SAR) can well overcome these shortcomings of optical imaging because of SAR working in an all-weather environment. Due to the imaging mechanism of SAR image is essentially different from optical image, the interpretation of SAR image is a huge challenge. Inspired by the powerful image-to-image translation capability of Generative Adversarial Networks (GANs), this paper proposes an improved Pix2Pix network to achieve the translation task from SAR image to optical image. In order to better maintain the structural information after image translation, this paper introduces a constraint based on phase consistency as the consideration of the loss function. Experimental results show that the proposed method has obvious advantages for multimodal remote sensing data translation tasks in comparison with the current state-of-the-art methods.
What problem does this paper attempt to address?