Land use/land cover classification for Göktürk-2 satellite

In this study, land-use/land-cover method for Gokturk-2 satellite was developed. Moreover, the developed method can be adapted to any satellite. The aim is to classify pixels in images captured by Gokturk-2 into four basic categories (soil, green areas, water bodies, and others), and to form a persistent mathematical model for this purpose. “Partial Least Squares for Discriminant Analysis (PLS-DA)” method is employed. This method forms a new sample space in such a way that it maximizes the covariance between samples and observations (in this study, each observation represents the class which the corresponding sample belongs to). Axes for the new sample space represent four classes rather than four physical bands (red, green, blue, and NIR) Gokturk-2 has. In this new space; two normal distributions from training data (one for class samples, and one for non-class samples) for each class is calculated in order to find an acceptance threshold. These thresholds are used in classification process. Test samples picked from the test dataset (three satellite images) show that the developed method reached average accuracy rate of 96.51%; while the Maximum Likelihood Estimator, which is frequently used in land use/land cover classification, (also included in commercial products such as ENVI, ERDAS) attained 86.13%.