Towards Faithful Reasoning in Remote Sensing: A Perceptually-Grounded GeoSpatial Chain-of-Thought for Vision-Language Models

TL;DR

The paper tackles the lack of verifiable reasoning in Vision-Language Models for remote sensing by introducing Perceptually-Grounded Geospatial Chain-of-Thought (Geo-CoT). It presents a two-stage training pipeline (Stage I: supervised fine-tuning on a large Geo-CoT380k corpus, Stage II: Group Relative Policy Optimization) to instill a cognitive architecture and refine factual correctness, yielding RSThinker which outputs both answers and a verifiable reasoning trace. The approach achieves state-of-the-art results across fine-grained perception, holistic understanding, and complex geospatial reasoning tasks, supported by extensive ablations and qualitative analyses. Public release of the Geo-CoT380k dataset and RSThinker provides a concrete path toward transparent, auditable reasoning in Earth Observation applications.

Abstract

Vision-Language Models (VLMs) in remote sensing often fail at complex analytical tasks, a limitation stemming from their end-to-end training paradigm that bypasses crucial reasoning steps and leads to unverifiable outputs. To address this limitation, we introduce the Perceptually-Grounded Geospatial Chain-of-Thought (Geo-CoT), a framework that models remote sensing analysis as a verifiable, multi-step process. We instill this analytical process through a two-stage alignment strategy, leveraging Geo-CoT380k, the first large-scale dataset of structured Geo-CoT rationales. This strategy first employs supervised fine-tuning (SFT) to instill the foundational cognitive architecture, then leverages Group Reward Policy Optimization (GRPO) to refine the model's reasoning policy towards factual correctness. The resulting model, RSThinker, outputs both a final answer and its justifying, verifiable analytical trace. This capability yields dominant performance, significantly outperforming state-of-the-art models across a comprehensive range of tasks. The public release of our Geo-CoT380k dataset and RSThinker model upon publication serves as a concrete pathway from opaque perception towards structured, verifiable reasoning for Earth Observation.

Figures (15)

• Figure 1: An overview of the RSThinker framework. Our novel Geo-CoT380k dataset (a) enables our two-stage alignment strategy (b) to instill a verifiable reasoning process (c), yielding state-of-the-art performance across a comprehensive suite of benchmarks (d).
• Figure 2: The two-stage alignment process. Our training strategy first instills a foundational cognitive architecture via supervised fine-tuning (SFT) and then refines this architecture's faithfulness via outcome-based reinforcement learning (GRPO).
• Figure 3: Comparison of RSThinker with SOTA VLMs on Object Detection task.
• Figure 4: Comparison of RSThinker with existing generic and RS VLMs on Visual Grounding task.
• Figure 5: Qualitative example of RSThinker's Geo-CoT: a methodical Planning-Grounding-Synthesis sequence culminating in a justified <answer>.