A study of pixel level fusion techniques for SAR and optical images

Abstract

Remote sensing (RS) imagery has been widely used in various applications such as detecting the land use and land cover, observing climate changes, and controlling forest fires. The remote sensing technology can be divided into two categories: optical remote sensing and polarimetric synthetic aperture radar (PolSAR). However, there is a common problem persistently impede the land surface analysis for optical remote sensing, which is the cloud cover. As thick cloud can partially or completely obstruct the scene and corrupt the optical imagery. The removal of cloud for optical remote sensing imagery has become a serious hindrance. Therefore, the purpose of this project is to reconstruct the cloud-contaminated region of optical images by using PolSAR images as additional information. In this report, the author presents various approaches to achieve the reconstruction of cloud-contaminated optical images. Mainly consists of two approaches: Traditional Pixel-Pixel Replacement Method: Making use of similarity algorithms and circle searching method to optimize the performance. Residual Neural Network Method: One of the famous deep learning models which can be used to learn the features of cloud-contaminated images, cloud-free images, and PolSAR images, so that it can predict the cloud-free images.