Sentinel-2 Image Scene Classification over Alentejo Region Farmland

Abstract

Given the wide-ranging farmland area, optical satellite images of farms are used to develop maps that reflect land dynamics and its behavior over different time frames, crops, and regions on various environmental conditions. In this regard, it is essential to identify and remove atmospheric distorted images to further prevent misleading information, since their presence severely restrict the use of optical satellite images for forecasting harvest dates, yield estimation, and manufacturing control in agriculture systems. These atmospheric distortions are frequent due to cloud, shadow, snow, and water cover over farmland. In this work, we developed a method to identify distortion covering images of corn crop farmland situated in the Alentejo Region of Portugal. The results are compared with the state-of-the-art (SOTA) Sen2Cor algorithm of the European Space Agency. Further, experimental results show that the developed image scene classifier model outperforms Sen2Cor by 10% in F1-measure.