.png){kind=icon}
.png){kind=icon}
Return
SLFNet: an improved boundary-sensitive multi-tasks deep network for agricultural parcel delineation using high-resolution remotely sensed imagery
.png){kind=icon}
PRE
AI
K
P
Y
Z
DOI：10.1080/17538947.2026.2632409.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Accurate delineation of agricultural parcels is essential for precision agriculture, global food security and sustainable land management within the Digital Earth framework. However, existing methods often exhibit limited global contextual awareness, inadequate boundary refinement, and weak adaptability to diverse landscapes. To address these challenges, we propose SLFNet, a boundary-sensitive multitask deep network for agricultural parcel delineation using high-resolution remote sensing imagery. SLFNet integrates three key innovations: (1) a Simplified Transformer-based Module (STM) for capturing long-range dependencies; (2) a Level-Aware Dilated Convolution and Shuffle (LDCS) module for adaptive multi-scale feature extraction; and (3) a Figure-of-Merit contour Loss (FOM Loss) to enhance boundary localization. SLFNet was evaluated on three geographically distinct regions (Denmark, Xinjiang, Sanyuan) with diverse heterogeneity using multiple high-resolution datasets. Experimental results indicate that SLFNet demonstrates competitive performance compared with several state-of-the-art baselines (e.g., HBGNet and REAUNet). Transferability experiments under limited target-domain supervision show that SLFNet can effectively adapt to fragmented agricultural landscapes, even with a small number of labeled samples. SLFNet provides a robust boundary-aware framework for agricultural parcel delineation and facilitates the extraction of structured agricultural information from high-resolution remote sensing imagery within the Digital Earth context.
Keywords:
Deep learning
agricultural parcel delineation
high-resolution remote sensing imagery
multitask learning
boundary-aware segmentation
Journal
IF:
4.9
Papers:
1.9K
Citations:
4.7K
Organization
Cited Papers