Satellite image fusion for land cover classificstion

Abstract

Synthetic Aperture Radar is one of the most widely used systems in modern radar technology because of its ability to analyze land-mapping information. SAR could produce images without affecting by objects such as clouds and haze, but the resolution of these images are limited. However, optical images can generate high-resolution images. Therefore, to obtain better land cover classification, this project will conduct image fusion for SAR image and optical image; and display the optimized land cover classification result. Firstly, in this report, multiple image filtering methods including spatial filtering, Wiener filtering and morphological filtering will be compared. Among them, morphological filtering shows the best performance for SAR image filtering objective. Secondly, different image segmentation methods including watershed segmentation, LAB segmentation, Chan-Vese segmentation and K-means segmentation will be discussed. All methods show their advantages and disadvantages in separating regions of SAR image. For image covering Singapore and Malaysia, K-means method is recommended. Lastly, image fusion result for segmented SAR image and optical image is analyzed by setting segments to half transparent. By implementing all these image processing functions, image fusion for land cover classification is achieved.