Related Experiment Videos
A 10-m National Agricultural Parcel Dataset of Ukraine Generated with an Enhanced Multi-Task Network and Sentinel-2
Junbin Li1,2, Hongwei Zeng3,4, Hang Zhao1,2,5
1State Key Laboratory of Remote Sensing and Digital Earth, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing, 100101, China.
None:
Agricultural parcels (APs) are fundamental spatial units for precision agriculture and effective agricultural management. However, large-scale AP mapping remains limited, particularly at national scales. To address this gap, this study proposes a national-scale AP mapping framework based on 10-m resolution Sentinel-2 imagery by enhancing the Hierarchical Boundary-Guided multi-task Network (HBGNet). Specifically, we developed HBGNet 2.0, an upgraded version of HBGNet, which introduces a novel self-adaptive weighting loss function, thereby balancing the learning of different tasks and improving AP delineation. Using HBGNet 2.0, we generated a 10-m resolution national-scale AP dataset for Ukraine in 2023. HBGNet 2.0 demonstrates the best overall performance among all compared models (ResUNet-a, BFINet, BsiNet, SEANet, and HBGNet), achieving the highest OA (92.86%), F1 (91.03%), and IoU (85.70%) metrics. Compared with three publicly available national AP datasets for Ukraine, our dataset exhibits better spatial detail and higher delineation accuracy. This high-precision AP dataset has the potential to serve as a critical resource for Ukrainian agricultural monitoring and for assessing conflict-related crop damage.
