Investigating the Segment Anything Foundation Model for Mapping Smallholder Agriculture Field Boundaries Without Training Labels

TL;DR

This work addresses the challenge of mapping smallholder agricultural field boundaries in data-scarce regions by evaluating the zero-shot Segment Anything Model (SAM) on 2 m SkySat imagery from Bihar, India. It systematically tests SAM across checkpoints, input sizes, multi-date imagery, edge-enhancement, and prediction fusion to gauge boundary delineation without any training. The results show SAM can detect about $58 ext{ %}$ of boundaries when combining original and edge-enhanced inputs, with multi-date fusion delivering the largest gains (up to $0.50$ detection) and mean delineation near $0.69$ IoU in favorable configurations. This work demonstrates a proof-of-concept for applying foundation models to geospatial boundary mapping in data-scarce contexts and highlights practical pathways for GeoXAI to support agricultural analytics and training-data generation.

Abstract

Accurate mapping of agricultural field boundaries is crucial for enhancing outcomes like precision agriculture, crop monitoring, and yield estimation. However, extracting these boundaries from satellite images is challenging, especially for smallholder farms and data-scarce environments. This study explores the Segment Anything Model (SAM) to delineate agricultural field boundaries in Bihar, India, using 2-meter resolution SkySat imagery without additional training. We evaluate SAM's performance across three model checkpoints, various input sizes, multi-date satellite images, and edge-enhanced imagery. Our results show that SAM correctly identifies about 58% of field boundaries, comparable to other approaches requiring extensive training data. Using different input image sizes improves accuracy, with the most significant improvement observed when using multi-date satellite images. This work establishes proof of concept for using SAM and maximizing its potential in agricultural field boundary mapping. Our work highlights SAM's potential in delineating agriculture field boundary in training-data scarce settings to enable a wide range of agriculture related analysis.

Figures (13)

• Figure 1: SkySat 2 m resolution satellite images for Bihar, India for dates, (d) 11 May 2015, (e) 29 September 2015, (f) 16 November 2015 and (g) 10 Feb 2016.
• Figure 2: Schematic diagram of the different levels of accuracy assessment for predicted layers.
• Figure 3: Detection and delineation accuracy at level 1 (raw predictions) and level 2 (after combining predictions from different checkpoints). Different bars represent different checkpoints. The Y axis represents detection accuracy, and the X axis shows bars grouped by input image chip size. The text on top of each bar shows the mean IoU value for that group.
• Figure 4: Accuracy metrics at level 2 (different image dates after combining checkpoints) and level 3 (combining image sizes).
• Figure 5: Accuracy metrics for original and edge enhanced images at different level of accuracy assessment.