Graph-based method based on Gaussian Mixture Modeling to classify agricultural lands

In this study, a graph-based method based on Gaussian Mixture Modeling (GMM) to classify agricultural products is proposed. The effects of different number of components and smoothing constants to the classification accuracies are investigated during the analyses. Tests are performed over two 4-channel Kompsat-2 satellite images which cover approximately 100 km² of the Karacabey Plain of the city Bursa. Based on the results computed, the proposed graph-based approach reached overall accuracies of 69% and 91,8% for the June and July Kompsat-2 images, respectively. Besides, it is observed that the graph-based approach improves the overall accuracies of around 10% compared to the results of the GMM classification which ignores the relations between neighboring pixels.