Determining and labeling the spectral signatures of land use / land cover classes using spectral indexes

Today, the rapid development of remote sensing technologies has led users to seek ways to develop robust and effective alternatives for data analysis. One of the classification approaches which is used to create land classes from satellite images is uncontrolled classification method. Uncontrolled classification approach is a more useful than controlled classification methods since it does not require prior knowledge of the studied area, moreover, it works without training data. To produce results, however, this approach requires different parameters such as number of classes, learning rate, number of iterations. Besides, since training data is not used in uncontrolled classification method, the accuracy rate of the results provided by this method is generally lower than the controlled classification. The earth studies success is in a close relationship with accuracy and reliability of the used information. Additionally, quick and economic obtainment of the aforementioned information is an important demand. In this study, a pilot study was carried out to create a spectral signature library in order to automatically identify the most used land cover classes on the image by using multispectral images of the Sentinel-2 satellite. In this direction, Sentinel-2 images of January, April, July and October belonging to seven regions of Turkey were examined. Spectral indices were calculated on these images and value ranges that distinguish land cover classes were determined. As a result of these processes, a spectral index value range library was tried to be created in the land cover classes determined in the images of the Sentinel-2 satellite and it was aimed that these value ranges constitute a base for automatic labeling processes.