Multi-Region Transfer Learning for Segmentation of Crop Field Boundaries in Satellite Images with Limited Labels

TL;DR

This work tackles crop field boundary delineation in satellite imagery under label scarcity by introducing multi-region transfer learning, pre-training on a high-quality region, fine-tuning with bridging data from another region, and evaluating on a third target region. It combines a Spatio-Temporal U-net (ST-U-net) backbone with multi-temporal inputs to predict border and interior masks, demonstrating substantial improvements over baselines across France, South Africa, and Kenya. The approach leverages freely available data for pre-training and targeted, smaller datasets for fine-tuning, and demonstrates that performance gains are especially pronounced when transitioning to data-poor regions. By providing open datasets and code, the authors enable practical deployment and benchmarking in diverse agricultural landscapes and contribute to scalable monitoring in global food security contexts.

Abstract

The goal of field boundary delineation is to predict the polygonal boundaries and interiors of individual crop fields in overhead remotely sensed images (e.g., from satellites or drones). Automatic delineation of field boundaries is a necessary task for many real-world use cases in agriculture, such as estimating cultivated area in a region or predicting end-of-season yield in a field. Field boundary delineation can be framed as an instance segmentation problem, but presents unique research challenges compared to traditional computer vision datasets used for instance segmentation. The practical applicability of previous work is also limited by the assumption that a sufficiently-large labeled dataset is available where field boundary delineation models will be applied, which is not the reality for most regions (especially under-resourced regions such as Sub-Saharan Africa). We present an approach for segmentation of crop field boundaries in satellite images in regions lacking labeled data that uses multi-region transfer learning to adapt model weights for the target region. We show that our approach outperforms existing methods and that multi-region transfer learning substantially boosts performance for multiple model architectures. Our implementation and datasets are publicly available to enable use of the approach by end-users and serve as a benchmark for future work.

Figures (4)

• Figure 1: Comparison between instance segmentation of crop fields in satellite remote sensing images (top) and example image from COCO dataset (bottom). In the top row, $t_0$, $t_1$, and $t_n$ represent satellite images acquired at different dates during the year. Top image shows field boundary instances in transparent white and bottom image shows object instance (person) in transparent yellow.
• Figure 2: Illustration of multi-region transfer learning approach. Model architecture adapted from aung2020farm.
• Figure 3: Example images and ground-truth masks from Sentinel-2 and PlanetScope images from France dataset.
• Figure 4: Qualitative field boundary segmentation results for example test images in selected experiments from Table \ref{['tab:results']}.