DBPapers
DOI: 10.5593/sgem2017/23/S10.004

ANALYSIS OF LAND USE LAND COVER CLASSIFICATION RESULTS DERIVED FROM SENTINEL-2 IMAGE

A.M. MARANGOZ, A. SEKERTEKIN, H.AKCIN
Monday 11 September 2017 by Libadmin2017

References: 17th International Multidisciplinary Scientific GeoConference SGEM 2017, www.sgem.org, SGEM2017 Conference Proceedings, ISBN 978-619-7408-03-4 / ISSN 1314-2704, 29 June - 5 July, 2017, Vol. 17, Issue 23, 25-32 pp, DOI: 10.5593/sgem2017/23/S10.004

ABSTRACT

In this study, object-based Land Use Land Cover (LULC) classification performance of Sentinel-2 image has been tested by comparing other medium resolution satellite dataset of Zonguldak test field. The test field covering a small area around Zonguldak is located in the Western Black Sea region of Turkey. It is noted for being one of the main coal mining areas in the world. For the purpose of the study, pan-sharpened Landsat 8 image was used because of its nearly similar ground sampling distance (GSD). The RGB and NIR bands of Sentinel-2 were used for classification and comparison. As a first step, Landsat-8 pan-sharpened image was created using High Pass Filtering (HPF) pan sharp algorithm in ERDAS software package. Following this, resulted images were handled by the eCognition v4.0.6 software with the main steps of segmentation and classification. After determining the optimal segmentation parameters correctly, classification of main Land use/Land cover classes were compared with by Landsat-8 derived LULC classes. Furthermore, the results were verified visually using high resolution satellite image Worldview-2. The accuracy assessment as Kappa statistics for Sentinel-2 and Landsat-8 are 0.74 and 0.66, respectively. The obtained results revealed that Sentinel-2 LULC images give better results than Landsat-8.

Keywords: Sentinel-2, Landsat-8, Land Use, Land Cover Image Contents, eCognition, Segmentation, Object-Based Image Analysis (OBIA)