An Assessment of Object-based Classification Compared to Pixel-based Classification in Google Earth Engine Using Random Forest

Abstract

Land use is one of the dynamic features that has an impact on environmental conditions. As the study area, the coastal area in the City of Cilegon, Province of Banten is subjected to land use dynamics for its economic development. Accordingly, this study aimed to provide the land use/land cover (LULC) classification within the study area in the year of 2021. The classification was done using Sentinel-2 images and processed on a free, open-access Google Earth Engine (GEE) environment. In generating the LULC classification, this study applied two approaches, i.e., Object-based Classification (OBC) and Pixel-based Classification (PBC), in order to get a better result in providing the LULC data. The predictor variables integrated several spectral indices and bands from the Sentinel-2. For the OBC, image segmentation was performed with a Simple Non-Iterative Clustering (SNIC). And, the classifier used for the OBC and PBC was Random Forest (RF). As a result, the study area consists of heterogeneous landscape including agricultural area, industrial area, settlement and other vegetated areas. Based on the accuracy assessment, the OBC outperformed the PBC with an overall accuracy at 0.95 and 0.731, respectively.