Cross-modality learning for earth surface mapping with cloud-covered satellite optical images and radar images

There are two common kinds of images used in land classification and recognition in remote sensing technology: optical images and polarimetric synthetic aperture radar (PolSAR) images. However, optical images can be covered by clouds for a long time due to weather problems, which is one of the most...

Full description

Bibliographic Details
Main Author: Pi, Ziyi
Other Authors: LU Yilong
Format: Final Year Project (FYP)
Language:English
Published: Nanyang Technological University 2021
Subjects:
Online Access:https://hdl.handle.net/10356/151026
Description
Summary:There are two common kinds of images used in land classification and recognition in remote sensing technology: optical images and polarimetric synthetic aperture radar (PolSAR) images. However, optical images can be covered by clouds for a long time due to weather problems, which is one of the most serious challenges for remote sensing technology. Therefore, the purpose of the project is to recover the landscape for cloud-covered areas on optical images, based on the reference from PolSAR images. In the report, the author includes different kinds of approaches to recover the landscape, including three following methods to derive the optimal method: Direct Method: Different kinds of searching or similarity algorithms. Poisson Method: Seamless cloning or mixing gradients. Pan Sharpening Method: Combination of Poisson and Pan Sharpening.