Weakly Supervised Spatio-Temporal Candidate Discovery of Dairy Farm Sites from Seasonal Satellite Imagery

2026-07-14

AI Summary

This study addresses the scarcity of labeled dairy farm annotations in satellite imagery by proposing an unsupervised candidate region discovery method that integrates multi-season Sentinel imagery with open-map priors. The approach employs a Barlow Twins encoder to learn season-invariant representations and combines vegetation indices, building indices, and pasture access patterns into a rule-based scoring system. Spatial smoothing is further applied on a geo-semantic graph to enhance localization consistency. Without requiring precise farm-level labels, the method identifies 535 high-scoring tiles from a dataset of 26,722 image tiles, which are then clustered into 71 candidate regions. The top five clusters achieve localization accuracies of 0.60 and 0.80 within 500-meter and 1000-meter radii, respectively, significantly improving the efficiency and reliability of livestock facility detection under weakly supervised conditions.

Abstract

Farm site discovery from satellite imagery is a spatiotemporal candidate ranking problem because farm evidence is distributed across pasture, field boundaries, roads, buildings, and seasonal vegetation patterns. Direct farm labels are often incomplete, which makes fully supervised detection difficult. This paper proposes a weakly supervised pipeline for ranking dairy farm candidate clusters from seasonal Sentinel imagery and open map priors. The method uses aligned spring, summer, and autumn image tiles from County Cork, Ireland, with spectral bands, vegetation indices, built area indices, and a pasture channel. A Barlow Twins encoder learns multi-season tile embeddings without farm labels. In parallel, weak OpenStreetMap farm priors are split into a prior and a held-out set. Prior features support a rule-based tile score that combines farm proximity, seasonal pasture evidence, and summer greenness, while held-out features are reserved only for proxy evaluation. The rule score is smoothed over a spatial representation graph using geographic proximity and embedding similarity, and high-scoring tiles are grouped into ranked candidate clusters. From 26,722 valid tiles, the main run selects 535 high-confidence tiles and forms 71 candidate clusters. The top 5 clusters achieve 0.60 precision within 500 m and 0.80 precision within 1000 m of held-out OpenStreetMap farm features. The top 10 clusters achieve 0.40 precision within 500 m and 0.80 precision within 1000 m. The results show that seasonal representation learning and weak geographic priors can reduce large satellite image collections into compact candidate sets for human review.

Problem

• weakly supervised
• spatio-temporal candidate discovery
• dairy farm detection
• seasonal satellite imagery
• candidate ranking

Innovation

• weakly supervised learning
• spatio-temporal representation
• seasonal satellite imagery
• Barlow Twins
• candidate cluster ranking