Deep Learning-Based Multi-Sensor Remote Sensing for Automated Jute Crop Mapping

Jute, India’s second most important natural fibre, holds significant historical and economic importance globally, leading to increased demand. In the Murshidabad district of West Bengal, which represents over $10 \%$ of India’s jute cultivation area, jute farming plays a vital role in the local economy and agricultural landscape. This paper utilizes multi-sensor data from Sentinel-1 and Sentinel-2, collected for the jute crop cycle for jute mapping. Deep learning-based segmentation models are implemented for automated jute crop mapping, with the UNet model performing best, achieving a precision (0.94), recall (0.85), IoU (0.817) and dice score (0.827), indicating accurate mapping of the jute crop. Further improvements were made using data augmentation techniques, hyperparameter tuning, and optimization.