Improved LULC Mapping by Combining Pixel-based and Object-based Fuzzy Classification Using Spectral Indices of RapidEye and GRASS GIS for Lao Cai area, Vietnam

Presenter : Pavithra Jayasinghe

Type: Poster

In this study, a combined of pixel-based and object-based fuzzy approach for land use/ land cover (LULC) classification from RapidEye imagery to improve classification accuracy is presented. The study area is Lao Cai – the mountainous province in north of Vietnam where the main of LULC classes are agriculture, forest, water, built-up and soil. The methodology of the combination comprises three main steps. The first step is fuzzy pixel-based classification, utilizes Normalized Difference Vegetation Index (NDVI) and Normalized Difference Water Index (NDWI) to distinguish water, non-water and vegetation classes. Second step is non-water classification. A threshold of 0.15 has been applied for segmentation, and then object-based fuzzy classification uses Soil index (SI) which extracted from Vegetation-Soil-Water index (VSW). Soil index (SI) makes it possible to distinguish water, soil, built-up, agriculture and vegetation classes from non-water class. Third step is vegetation classification. Vegetation class extracted from fuzzy pixel-based classification has been merged with vegetation class extracted non-water classification. A threshold of 0.25 has been used for segmentation, and then object-based fuzzy classification has been applied for Normalized difference Red Edge index (NDRE), NDRE/SI ratio and Digital Elevation Model (DEM) of the class to separate agriculture and forest classes. The workflow of the study has been implemented using GRASS GIS. The accuracy assessment have been carried out with reference data are indicating strong correlation with combined fuzzy classification results. An improved classification results have reported (85.7 %) while comparing with the results of pixel based (82.2 %) and object based classification (78.5 %) individually.