Satellite optical and radar image fusion for land cover classification

Abstract

Nowadays, Synthetic Aperture Radar has become the most widely used radar system, since SAR images are seldom affected by weather conditions such as fog, clouds and haze. On the other hand, optical image captures more details of target object by detecting reflected solar EM radiation. However, both SAR and optical images have their own limitation. The visible details of SAR images are limited, because typical radar frequency is out of visible light frequency range. Optical images could be affected by nighttime and weather factor such as cloud and haze. Therefore, both SAR and optical images need to be combined to achieve a better land cover structure analysis result. The aim of this project was to study on image processing techniques, which will combine both SAR and optical images together to obtain a better land cover structure analysis result.

Firstly, SAR images will be analyzed and processed with different filtering methods which includes wiener and morphological filtering method.

Following that, different segmentation techniques will be applied to the filtered SAR image, such as Watersheld, Chan-Vese, Superpixel and K-Means segmentation methods.

Thirdly, SAR segmentation result and optical images will be registered with two image registration techniques including Feature points manual selection registration and Mutual information based image registration.

Finally, registered SAR segmentation image and optical image were combined through image fusion techniques to achieve a better monitoring of land cover classification.