Image translation in optical-to-SAR remote sensing images with GAN

• Main Author: Liao, Xuan Li
• Other Authors: Wen Bihan
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/167216
• Institution: Nanyang Technological University

In recent years, Synthetic Aperture Radar (SAR) imagery has emerged as a significant tool in the domain of Earth observation and object detection. SAR images possess several advantages over traditional optical satellites, owing to their utilization of microwaves. However, the current amount of SAR images is not sufficient with the optical images. Image translation improves a lot through the development of Generative Adversarial Networks (GAN). However, there is a lack of application in the remote sensing domain due to the shortness of the high-resolution paired opt-to-SAR datasets. In this project, paired and unpaired GAN-based methods, Pix2Pix and CycleGAN are implemented on the newly proposed datasets and compared their performances for optical-to-SAR image translation. A novel high-resolution and quad-polarization dataset, SN6_EO_SAR, is introduced for optical-to-SAR image-to-image translation and the implementations of GAN-based method discussed in this project have been made available to facilitate reproducible research in the remote sensing field. Several quantitative image qualities are applied to evaluate the quality of the generated images. In addition, different resolution and polarization mode datasets were put together to make a comparison and investigate how the self-characteristic of the SAR images affects the quality of the images. In future work, various paired and unpaired GAN-based methods are applied to evaluate and compare the quality of the generated images. Potential applications of synthetic SAR images can be investigated. This could involve exploring the use of synthetic SAR images in various tasks such as target detection, terrain mapping, and change detection.