GeoReason: Bridging Logical Reasoning and Spatial Fidelity in Remote Sensing Segmentation

📄 arXiv: 2610.23416v1 📥 PDF

作者: Cheng Fan, Han Wu, Tingzhang Luo, Minjing Dong, Yehui Tang, Jianyuan Guo

分类: cs.CV

发布日期: 2026-10-16

🔗 代码/项目: GITHUB

💡 一句话要点

提出GeoReason以解决遥感分割中的逻辑推理与空间保真问题

🎯 匹配领域: 支柱九：具身大模型 (Embodied Foundation Models)

关键词: 遥感分割 逻辑推理 空间保真 多模态学习 深度学习 图像处理 机器学习

📋 核心要点

  1. 现有方法在复杂地理背景下进行遥感推理分割时，面临空间保真度下降和目标误识别的挑战。
  2. 本文提出GeoReason框架，通过将分割标记直接注入用户查询，防止信息衰减，并通过位置感知的思维链增强视觉推理。
  3. GeoReason在Earthreason和RRSIS-D基准测试中表现优异，超越了当前最先进的方法，显示出显著的性能提升。

📝 摘要（中文）

多模态大型语言模型（MLLMs）在指令驱动的感知方面取得了显著进展，但在复杂地理环境和极端尺度变化下，遥感（RS）推理分割的扩展仍然面临巨大挑战。本文识别出一个关键瓶颈：在自回归生成过程中，分割标记[SEG]与其初始视觉基础的顺序疏离，导致信息衰减，尤其对像素足迹微小的RS目标影响严重。为此，本文提出了GeoReason框架，通过引入预期先验机制和位置感知的思维链，增强逻辑推理并保持空间完整性。实验结果表明，GeoReason在Earthreason和RRSIS-D基准测试中表现优于现有最先进方法。

🔬 方法详解

问题定义：本文旨在解决遥感分割中逻辑推理与空间保真度之间的矛盾，现有方法在长推理链中导致分割标记信息衰减，影响目标定位准确性。

核心思路：提出GeoReason框架，通过将分割标记[SEG]作为初始条件而非终端输出，保持其作为空间锚点，防止信息在推理过程中的衰减。

技术框架：GeoReason框架包括两个主要模块：预期先验机制和位置感知思维链。前者将[SEG]标记注入用户查询，后者通过边界框实现粗到细的定位。

关键创新：最重要的创新在于将分割标记的生成过程从推理链的末尾转移到初始阶段，显著减少了信息衰减的风险。

关键设计：设计了位置感知的思维链以增强视觉推理，并引入潜在补丁到标记对齐损失（PTA Loss），确保分割标记与实际RS纹理的对齐。具体损失函数和网络结构的细节在实验部分进行了详细描述。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

在Earthreason和RRSIS-D基准测试中，GeoReason框架的性能显著优于现有最先进的方法，具体提升幅度达到XX%（具体数据需根据实验结果填写），展示了其在复杂遥感场景下的有效性和可靠性。

🎯 应用场景

GeoReason框架在遥感图像分析、环境监测和城市规划等领域具有广泛的应用潜力。通过提高遥感分割的准确性和可靠性，能够为决策支持系统提供更为精准的数据基础，推动智能城市和可持续发展目标的实现。

📄 摘要（原文）

Multimodal Large Language Models (MLLMs) have significantly advanced instruction-driven perception through the "Embedding-as-Mask" paradigm. However, extending this capability to Remote Sensing (RS) reasoning segmentation remains exceptionally challenging under cluttered geographical contexts and extreme scale variations, often leading to Spatial Fidelity Degradation and Target Misidentification. We identify a critical yet underexplored bottleneck: the sequential alienation of the segmentation token [SEG] from its initial visual grounding within the autoregressive generation paradigm. Produced only at the end of long reasoning chains, [SEG] relies on visual evidence that has been progressively diluted by linguistic abstraction and autoregressive noise. This information decay is particularly catastrophic for RS targets with minuscule pixel footprints, where even marginal fidelity loss leads to localization failure. To address these bottlenecks, we propose GeoReason, a novel framework designed to fortify logical deduction and preserve spatial integrity. First, we introduce an Anticipatory Prior mechanism that injects the [SEG] token directly into the initial user query, shifting it from a terminal output to a primary condition. This paradigm shift transforms the token into a persistent spatial anchor, preventing information decay during long-form linguistic inference. Second, we enhance visual reasoning via a Location-aware Chain-of-Thought (CoT) that enforces coarse-to-fine localization using bounding boxes, coupled with a Latent Patch-to-Token Alignment (PTA) Loss that explicitly aligns the latent segmentation token with physical RS textures. Extensive experiments demonstrate that GeoReason consistently outperforms state-of-the-art methods on the Earthreason and RRSIS-D benchmarks. Code is available at https://github.com/Fanc0126/GeoReason.