Think with Extra-Image: A Farmland Segmentation Agent Driven by Spatio-Temporal Information Gain

2026-07-30

AI Summary

Existing remote sensing-based agricultural field segmentation methods rely on single-temporal imagery, struggling to handle appearance ambiguities and information gaps caused by phenological variations and spatial context. This work reframes the task through the lens of the information bottleneck principle and introduces FarmSeeker—a novel agent-based dynamic segmentation architecture that enhances segmentation accuracy by adaptively retrieving external spatiotemporal information through an uncertainty-aware, query-driven strategy. To support dynamic inference evaluation, the study also constructs GSFS-Bench, the first global high-resolution benchmark dataset for farmland segmentation enabling temporal reasoning. Extensive experiments on this benchmark demonstrate that the proposed method achieves significant improvements in segmentation accuracy, stability, and robustness compared to existing approaches.

Abstract

Existing farmland remote sensing image (FRSI) segmentation follows a "Think with Intra-Image" paradigm, assuming that the current image contains sufficient visual evidence for reliable segmentation. Yet farmland appearance varies with phenology and spatial context and is often confused with other land-cover, making instantaneous, local observations inadequate. Thus, segmentation ambiguity stems not only from limited model representation, but more fundamentally from the required spatio-temporal information lying beyond the current image. Based on this insight, we redefine FRSI segmentation from an information bottleneck perspective as a dynamic decision process driven by task-relevant extra spatio-temporal information gain. We further propose FarmSeeker, a dynamic FRSI segmentation agent that identifies ambiguous regions, reasons about their causes, and queries extra spatio-temporal information on demand for accurate segmentation. To evaluate FarmSeeker, we construct GSFS-Bench, the first global-scale, high-resolution FRSI segmentation benchmark that supports reasoning-querying. Experiments show that FarmSeeker achieves more stable segmentation performance than existing methods. The project is publicly available at: https://withoutocean.github.io/FarmSeeker/

Problem

• farmland segmentation
• remote sensing
• spatio-temporal information
• segmentation ambiguity
• information bottleneck

Innovation

• spatio-temporal information gain
• dynamic segmentation agent
• farmland remote sensing
• information bottleneck
• reasoning-querying framework