Farmland Extent and Visible Boundary Mapping from 1 m NAIP Imagery Using Residual U-Net and Text-Prompted SAM 3 Refinement

2026-07-24 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Advanced, quick

Summary

A reproducible workflow maps farmland extent and visible boundaries using 1 m NAIP RGB imagery. The study utilized 37 diverse scenes, annotated in CVAT, yielding 5,698 256x256 image patches for training (3,850), validation (770), and testing (1,078). A Residual U-Net (ResUNet), trained with a Dice-dominant loss L = 2.5(1 - Dice) + BCE, achieved a test accuracy of 0.8808, IoU of 0.8605, and Dice score of 0.9234. This ResUNet was then fused via logical OR with a frozen SAM 3 branch, prompted by "agricultural farmland field." This fusion significantly improved Dice scores on challenging patches, from 0.858 to 0.955 for orchard rows and from 0.804 to 0.903 for fragmented parcels. The workflow employs sliding-window stitching to produce coherent regional masks, with example tile Dice scores of 0.898 and 0.919. The resulting semantic farmland-extent layer supports agricultural monitoring in areas lacking current field data.

Key takeaway

For agricultural analysts or GIS specialists needing current farmland maps, you should consider this reproducible workflow. It provides a robust method for generating semantic farmland-extent layers from 1 m NAIP imagery. Integrating a Residual U-Net with text-prompted SAM 3 significantly enhances boundary detection accuracy, particularly in challenging, fragmented, or peri-urban agricultural areas. This approach offers a valuable alternative when proprietary field data is incomplete or outdated.

Key insights

Combining a specialized U-Net with text-prompted SAM 3 effectively maps farmland boundaries from NAIP imagery, improving accuracy on complex terrains.

Principles

• Fusing specialized models enhances segmentation.
• Text prompts guide foundation models effectively.
• Dice-dominant loss improves boundary detection.

Method

Annotate 1m NAIP RGB imagery for farmland extent. Train ResUNet with Dice-dominant loss. Fuse ResUNet output with text-prompted SAM 3 via logical OR. Apply sliding-window stitching for regional masks.

In practice

• Annotate imagery in CVAT for binary masks.
• Apply ResUNet for initial farmland segmentation.
• Refine boundaries using text-prompted SAM 3.

Topics

• Farmland Mapping
• NAIP Imagery
• Residual U-Net
• SAM 3
• Semantic Segmentation
• Agricultural Monitoring
• Remote Sensing

Best for: Computer Vision Engineer, AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Promptable Concept Segmentation from Above: Evaluating SAM 3's Zero-Shot and One-Shot Capabilities in Remote Sensing — SAM 3 shows promise in remote sensing but suffers from ground-level semantic bias and sub-pixel resolution limits.
• From Boundaries to Semantics: Prompt-Guided Multi-Task Learning for Petrographic Thin-section Segmentation — Petro-SAM enables high-quality joint grain-edge and lithology segmentation in petrographic images by adapting SAM.
• SWNet: A Cross-Spectral Network for Camouflaged Weed Detection — Cross-spectral imaging and boundary refinement improve camouflaged weed detection in agriculture.
• Context-Aware Slum Mapping in Sub-Saharan Africa Using Sentinel-1 Texture and Local Climate Zones — Integrating Sentinel-1 SAR textures with Sentinel-2 optical data significantly enhances informal settlement mapping accuracy and seasonal stability in Sub-Saharan Africa.
• Heterogeneous LiDAR Early Fusion and Learned Re-Ranking Strategy for Robust Long-Term Place Recognition in Unstructured Environments — Heterogeneous LiDAR early fusion and learned re-ranking significantly boost place recognition in challenging unstructured agricultural environments.
• Prompt-Calibrated SAM 3 for Open-Vocabulary Remote Sensing Semantic Segmentation — Calibrating SAM 3's prompt interface with offline construction, cached embeddings, and guided fusion significantly improves remote sensing OVSS.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.