Agribound

Agribound provides a unified interface for agricultural field boundary delineation using satellite imagery. It leverages geospatial foundation models, pre-trained segmentation, and embeddings to offer a comprehensive solution. This tool supports various satellite sources, including Google Earth Engine-based imagery like Landsat, Sentinel-2, HLS, NAIP, and SPOT, as well as local GeoTIFFs and pre-computed embedding datasets such as Google Satellite Embedding and TESSERA.

The pipeline within Agribound manages the entire process, from composite building and optional fine-tuning to the delineation engine, post-processing (smoothing, simplifying, filtering), and LULC crop filtering before final export. A key feature is its automatic removal of non-agricultural polygons, differentiating it from other packages that detect all visual boundaries. This functionality uses land-use/land-cover data, automatically selecting the best available dataset based on the study area, such as USGS Annual NLCD for the US or Google Dynamic World globally.

The tool is designed for ease of use, with installation via pip and a straightforward API for initiating delineation tasks. It returns a GeoDataFrame containing field boundary polygons along with attributes like area. Agribound demonstrates its capabilities through examples like supervised delineation using DINOv3 + SAM2 on NAIP imagery and unsupervised delineation with TESSERA + LULC Filter + SAM2 on Sentinel-2, showcasing its versatility across different regions and engines.