Land Cover Classification through Ontology Approach from Sentinel-2 Satellite Imagery

Abstract

This study proposed a land cover classification through an ontology approach from Sentinel-2 satellite imagery. Five steps were conducted as the research workflow, those were (1) Determine reference data and land cover classes; (2) Determine visual image interpretation features; (3) Feature extraction; (4) Ontology rules design; and (5) Land cover classification using ontology approach. The image feature of ontology was acquired by semantic reasoning approach from Indonesia National Standard RSNI-1 Land Cover Class in Medium Resolution Optical Imagery document. The features comprised of NDVI, Brightness, GLCM homogeneity and Rectangular fit. Image segmentation produced 2072 segments/objects by using eCognition software. Overall accuracy was used to evaluate the performance of the classification result. In addition, classification without ontology approach was carried out using CNN to compare the ontology result. Classification with ontology rule produced overall accuracy 99.8% whereas classification without ontology rules produced overall accuracy 98%.