Deep Learning in the Mapping of Agricultural Land Use Using Sentinel-2 Satellite Data

Continuous observation and management of agriculture are essential to estimate crop yield failure. Remote sensing is cost-effective, as well being an efficient solution monitor on a larger scale. With high-resolution satellite datasets, the monitoring mapping agricultural land easier more effective. Nowadays, applicability deep learning continuously increasing in numerous scientific domains due availability high-end computing facilities. In this study, (U-Net) has been implemented different use types over part Punjab, India, using Sentinel-2 data. As comparative analysis, well-known machine random forest (RF) tested. To assess land, major winter season types, i.e., wheat, berseem, mustard, other vegetation have considered. experimental outcomes, U-Net RF classifiers achieved 97.8% (kappa value: 0.9691) 96.2% (Kappa 0.9469), respectively. Since little information exists cultivated by smallholders region, study particularly helpful assessment mustard (Brassica nigra), berseem (Trifolium alexandrinum) acreage region. Deep remote data allows object-level detection earth’s surface imagery.