Accuracy and efficiency of compact foundation models for surface-water mapping during floods

📄 arXiv: 2610.34904v1 📥 PDF

作者: Parastoo Farajpoor, Mohammadreza Narimani

分类: eess.IV, cs.CV, cs.LG, stat.AP

发布日期: 2026-10-26

备注: 19 pages, 8 figures, 7 tables. Code: https://github.com/MohammadrezaNarimaniUCDavis/Prithvi_Tiny_Floodwater_GeoAI Data: https://doi.org/10.5281/zenodo.23653544

💡 一句话要点

评估紧凑型基础模型在洪水期间的水面映射效率与准确性

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

关键词: 水面映射 洪水监测 紧凑型模型 预训练 计算效率 深度学习 遥感技术

📋 核心要点

  1. 现有的水面映射方法在准确性和计算效率上存在不足，尤其是在洪水事件中。
  2. 本文提出了使用紧凑型预训练模型Prithvi-EO-2.0-Tiny进行水面映射，旨在提高效率并保持准确性。
  3. 实验结果表明，微调后的Prithvi-Tiny在计算速度上显著优于U-Net，且在特定条件下表现出竞争力的准确性。

📝 摘要（中文）

预训练的地球观测模型可能减少水面映射的工作量，但其价值取决于比较的替代方案。本文评估了Prithvi-EO-2.0-Tiny模型，使用了来自11次洪水事件的446个手工标注的Sentinel-2图像。通过与U-Net、较大的Prithvi模型和光谱指数阈值进行比较，实验考察了训练样本数量、模型拟合事件的排除以及计算成本和预测可靠性。结果显示，U-Net在252个训练样本下的水面交并比(IoU)为0.856，而微调后的Prithvi-Tiny为0.833。预训练对小型变换器的IoU提升为0.046，但未改变U-Net的排名。这些结果支持紧凑型预训练模型在计算需求低且准确性竞争的特定角色。

🔬 方法详解

问题定义：本文旨在解决洪水期间水面映射的准确性与计算效率问题。现有方法如U-Net在计算成本和处理速度上存在较大挑战，尤其是在处理大量图像时。

核心思路：论文提出通过使用紧凑型预训练模型Prithvi-EO-2.0-Tiny，结合微调策略，来提高水面映射的效率和准确性。该方法利用预训练的优势，减少了对大量标注数据的依赖。

技术框架：整体架构包括数据准备、模型训练与微调、性能评估等阶段。使用446个手工标注的Sentinel-2图像作为训练和测试数据，比较不同模型的表现。

关键创新：最重要的创新在于通过微调紧凑型预训练模型，尽管其在IoU上未能超越U-Net，但在计算效率上表现出显著优势，提供了一种新的思路来平衡准确性与效率。

关键设计：在实验中，模型的训练样本数量、事件排除策略、损失函数的选择等都经过精心设计。微调后的Prithvi-Tiny在GPU上处理512x512像素的图像仅需16.1毫秒，而U-Net则需30.6毫秒，显示出明显的速度优势。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

实验结果显示，在252个训练样本下，U-Net的水面交并比(IoU)为0.856，而微调后的Prithvi-Tiny为0.833。尽管Prithvi-Tiny在IoU上略逊一筹，但其在GPU上的处理速度为16.1毫秒，显著优于U-Net的30.6毫秒，展示了在计算效率上的优势。

🎯 应用场景

该研究的潜在应用领域包括洪水监测、环境保护和城市规划等。通过提高水面映射的效率和准确性，可以为灾害响应和资源管理提供更为及时和可靠的数据支持，进而提升应对自然灾害的能力。

📄 摘要（原文）

Pretrained Earth observation models may reduce the effort needed to map surface water, but their value depends on the alternative against which they are compared. We evaluated Prithvi-EO-2.0-Tiny using the 446 hand-labeled Sentinel-2 chips from 11 flood events in Sen1Floods11. A fine-tuned model, a frozen-encoder model, and an identical randomly initialized architecture were compared with a similarly sized U-Net, a larger Prithvi model, and spectral-index thresholds. Experiments varied the number of training chips, excluded events from model fitting, and measured computational cost and prediction reliability. With 252 training chips, U-Net achieved a water intersection-over-union (IoU) of 0.856 on the official test set, compared with 0.833 for fine-tuned Prithvi-Tiny. Pretraining improved the tiny transformer over random initialization by a reported 0.046 IoU with 25 training chips and 0.023 with 252 chips, but did not reverse the U-Net ranking. Across 11 event-exclusion evaluations, mean IoU was 0.768 for fine-tuned Prithvi-Tiny and 0.777 for U-Net; the paired interval included zero. Fine-tuned Prithvi-Tiny processed a 512x512-pixel chip in 16.1 ms on a GPU and 258 ms on a laptop CPU, compared with 30.6 and 665 ms for U-Net. Errors were concentrated near water boundaries, and high predictive entropy identified many, but not all, erroneous pixels. These results support a specific role for compact pretrained models: lower measured computational demand with competitive accuracy, rather than consistent superiority. Event-dependent performance, shared hyperparameter selection, and the distinction between observed water and newly inundated land delimit the conclusions.