A novel deep learning change detection approach for estimating spatiotemporal crop field variations from Sentinel-2 imagery

Abstract

The analysis of crop variation and the ability to quantify it is a critical and challenging task. Remote sensing (RS) has proven to be an effective tool for monitoring crops and detecting seasonal variations worldwide. This opens new opportunities for developing effective crop monitoring models, with deep learning models showing great promise. This study presents a deep learning-based U-Net v5 Change Detection (UCD) model capable of identifying and monitoring the spatio-temporal variations in crop fields. The application of the model is demonstrated using Sentinel-2 imagery over Patiala district in India to monitor the seasonal crop variation (rabi crop) during 2017–2018. The results have shown that the UCD model has achieved better results (95.6–98.4%) in accuracy for classified maps and more than (91.6%–96.6%) in accuracy for change maps. This study will be useful for crop monitoring, precision agriculture and crop yield prediction and can assist in decision and policy making towards a more sustainable environment. © 2024