Land cover classification using data fusion to support sustainable coffee certification processes

Abstract

Land cover classification is a remote sensing task that enables the visualization of different land uses. In the context of the sustainable coffee market, a land cover map is required as part of the sustainable coffee certification process. In this study, the land cover of a coffee production farm was classified into four categories: coffee, forest, civil infrastructure, and soil areas. Aerial images were acquired using a UAV equipped with visible and multispectral cameras. Image processing resulted in an orthomosaic for each camera, a vegetative index map (NDVI), and digital elevation models. Through statistical analysis and data fusion strategies, multithresholding and decision tree models-CART, Random Forest (RF), and Gradient Boosting (GB)-were trained and used to classify each pixel into one of the four categories. GB achieved the highest accuracy (94%), followed by RF (84%) and CART (83%). This study enhances the understanding of remote sensing methodologies and land use classification, specifically applied to the geographical particularities of the Colombian territory, and serves as a foundational step toward the application of agricultural technological innovation models in the country.