Cloud detection and information cloning technique for multi temporal satellite images

One of the main sources of noises in remote sensing satellite images are regional clouds and shadows of these clouds caused by atmospheric conditions. In many studies, these clouds and shadows are masked with multitemporal images taken from the same area to decrease effects of misclassification and deficiency in different image processing techniques, such as change detection and NDVI calculation. This problem is surpassed in many studies by mosaicking with different images obtained from different acquisition dates of the same region. The main step of all these studies that cover cloud cloning or cloud detection is the detection of clouds from a satellite image. In this study, clouds and shadow patches are classified by using a spectral feature based rule set created after segmentation process of Landsat 8 image. Not only spectral characteristics but also structural parameters like pattern, area and dimension are used to detect clouds and shadows. Information of cloud projection is used to strengthen cloud shadow classification. Rule set of classification is developed within a transferable approach to reach a scene independent solution. Results are tested with different satellite images from different areas to test transferability and compared to other state-of art methods in the literature. Detection of clouds and cloud shadows features correctly is the main step of cloning procedure to create cloudless image from multitemporal image dataset. Multitemporal image dataset is used to find best image to clone cloud image. Choosing best image for cloning process is an important step for reliable cloning. Statistical and seasonal similarity tests are used to find best image to clone cloud covered image. Vector intersections are used to find cloudless images between multitemporal dataset. Flood Fill method is used to create cloudless image from cloud covered image by using information extraction from cloudless images in dataset. Accuracy of cloning process is tested by using SSIM index to find structural and spectral similarity to cloudless image. All cloning results are tested with different image from different regions to check transferability of study. This study can be regarded as a scientific approach to create cloudless image mosaics for each kind of application. Method in this thesis is a scientific approach to well-known methods of famous cloudless mosaic generation methods of Google, Mapbox Co. etc. for creation of visually good-looking base maps for web maps.