An object-based approach for vegetation and non-vegetation discrimination using WorldView-2 image

Abstract

The emergence of remote sensing images with high spatial resolution has increased the advancement of image-based information extraction methods. One of the rapidly developing approaches for mapping and analyzing high spatial resolution images is the object-based approach, also known as geographic object-based image analysis (GEOBIA). This development makes it possible to quickly and accurately distinguish between vegetated and non-vegetated objects in vegetation study. This study aims to (1) create a ruleset to discriminate vegetated and non-vegetated objects from a high spatial resolution image, (2) apply the GEOBIA approach to map vegetated and non-vegetated objects, and (3) calculate the accuracy of the mapping results. The GEOBIA approach was applied to a WorldView-2 image (2 m pixel size and eight multispectral bands) of the Clungup Mangrove Conservation area, Malang, East Java, Indonesia. We assessed the ability of all of the WorldView-2 image bands for discriminating the targeted objects. The segmentation process in GEOBIA used a multi-resolution segmentation algorithm using the normalized difference vegetation index (NDVI), and the image classification used a rule-based classification technique. The green, red, and near-infrared bands can effectively distinguish the targeted objects based on the developed ruleset. The classification result shows that the vegetated and non-vegetated classes fall within their corresponding objects on the image. We implemented an area-based accuracy assessment that assesses both positional and thematic accuracy of the mapping result, based on the visual interpretation of the pansharpened WV-2 image (0.5 m pixel size) as a reference for the accuracy assessment. This process results in a 74,06% accuracy, meaning that the combination of GEOBIA and WorldView-2 image produces high accuracy of vegetated and non-vegetated objects map.