LAND COVER MAPPING IN CAMAU PROVINCE BY MACHINE LEARNING ALGORITHMS USING SENTINEL-2 IMAGERY

In this article, we have built a land cover map of Ca Mau province, Vietnam using 3 different classification methods random forest (RF), Support Vector Machine (SVM) and extreme gradient boosting (Xgboost). The study area is a mixed urban and rural area in the Mekong Delta, Vietnam with six land cover layers (LC). The satellite images used for classification are multi-temporal Sentinel-2 images from January to December 2021. The number of images in this period after cloud removal remains 17 images. The Median filtering method was used to generate an unique image in this time period for classification. The tool to do the classification is the Google Earth Engine platform. The sample is taken based on the land use map of Camau province in 2014 and Google Earth images. The number of samples taken for classification for all 3 methods is close to 4000 and the number of samples for the accuracy assessment is 3000 pixels. Overall (OA) error of SVM was 79.5, kappa coefficient was 0.72 while the Xgboost method achieved 85.6 and Kappa: 0.79 and FR was OA: 86.5 and Kappa: 0.81. The image classified by RF method was selected to build the map at 1:50,000 scale. © 43rd Asian Conference on Remote Sensing, ACRS 2022.