Agricultural crop classification using Sentinel-2 data and local features

Abstract

The satellite remote sensing data-based classification of agricultural crops is a critical tool for agricultural management. We propose a crop classification framework that combines complementary texture descriptors from Sentinel-2 multispectral images. The study identifies the most discriminant spectral bands and shows that the proposed method effectively distinguishes between various crops in the face of class imbalance issues. Our experiments demonstrate that random sampling strategy outperforms sequential sampling for crop classification. The resulting classification system enables precise automated crop inventories to facilitate agricultural resource planning and optimization. The method is of particular use in monitoring agricultural fields within regions experiencing the impacts of climate change and food insecurity.