Skip to main content
QUICK REVIEW

[论文解读] Building Extraction from Remote Sensing Images via an Uncertainty-Aware Network

Wei He, Jiepan Li|arXiv (Cornell University)|Jul 23, 2023
Automated Road and Building ExtractionEngineering被引用 10
一句话总结

论文提出 UANet,一种在高分辨率遥感图像中进行建筑提取的带不确定性感知的编码器-解码器框架,具备一个先验信息引导模块(PIGM)和一个不确定性感知融合模块(UAFM),由一个不确定性等级算法(URA)引导,以降低不确定性来 refinement 预测。

ABSTRACT

Building extraction aims to segment building pixels from remote sensing images and plays an essential role in many applications, such as city planning and urban dynamic monitoring. Over the past few years, deep learning methods with encoder-decoder architectures have achieved remarkable performance due to their powerful feature representation capability. Nevertheless, due to the varying scales and styles of buildings, conventional deep learning models always suffer from uncertain predictions and cannot accurately distinguish the complete footprints of the building from the complex distribution of ground objects, leading to a large degree of omission and commission. In this paper, we realize the importance of uncertain prediction and propose a novel and straightforward Uncertainty-Aware Network (UANet) to alleviate this problem. To verify the performance of our proposed UANet, we conduct extensive experiments on three public building datasets, including the WHU building dataset, the Massachusetts building dataset, and the Inria aerial image dataset. Results demonstrate that the proposed UANet outperforms other state-of-the-art algorithms by a large margin.

研究动机与目标

  • Motivate and address uncertain predictions in building extraction from high-resolution RS images.
  • Propose an end-to-end Uncertainty-Aware Network (UANet) that ranks and reduces pixel-wise uncertainty.
  • Develop a Prior Information Guide Module (PIGM) to enhance high-level features using prior extraction maps.
  • Introduce an Uncertainty-Aware Fusion Module (UAFM) and Uncertainty Rank Algorithm (URA) to refine features level-by-level and produce a low-uncertainty final map.
  • Validate performance across three public datasets and compare against state-of-the-art methods.

提出的方法

  • Use a conventional encoder–decoder backbone to obtain an initial uncertain extraction map.
  • Design the Prior Information Guide Module (PIGM) with cross-attention to leverage the initial map for spatial and channel guidance of the highest-level features.
  • Introduce the Uncertainty Rank Algorithm (URA) to quantify foreground and background uncertainty and produce rank-based weights.
  • Develop the Uncertainty-Aware Fusion Module (UAFM) to fuse high-level and lower-level features with uncertainty-guided weighting in a coarse-to-fine manner.
  • Train with a single BCE loss supervising multiple intermediate uncertainty maps (M1–M5) to optimize the final low-uncertainty extraction map.
Figure 2: The structure of the Uncertainty-Aware Network, which is composed of a general encoder-decoder, a prior information guide module (PIGM), and an uncertainty-aware fusion module (UAFM).
Figure 2: The structure of the Uncertainty-Aware Network, which is composed of a general encoder-decoder, a prior information guide module (PIGM), and an uncertainty-aware fusion module (UAFM).

实验结果

研究问题

  • RQ1Can uncertainty-aware modeling improve building footprint extraction in remote sensing images with diverse scales and backgrounds?
  • RQ2How can prior extraction maps be effectively used to guide high-level feature refinement in a cross-attention fashion?
  • RQ3Does incorporating foreground/background uncertainty ranking via URA improve decoding and final map certainty?
  • RQ4How does UANet perform compared with state-of-the-art methods on WHU, Massachusetts, and Inria building datasets?

主要发现

  • UANet achieves state-of-the-art performance across three public datasets, surpassing prior methods on IoU, F1, Precision, and Recall on WHU, Massachusetts, and Inria datasets.
  • Incorporating PIGM and cross-attention-based spatial/channel guidance improves high-level feature learning and final segmentation accuracy.
  • URA enables level-wise uncertainty-aware refinement, leading to a final extraction map with reduced uncertainty and sharper boundaries.
  • Compared to BuildFormer and other baselines, UANet shows notable gains in IoU (e.g., IoU improvements up to 1.84 percentage points on Inria) and better qualitative delineation of building footprints.
Figure 3: The structure of the Prior Information Guide Module (PIGM).
Figure 3: The structure of the Prior Information Guide Module (PIGM).

阅读本论文

在 Nubint 阅读器中打开原文,边读边就任意段落提问。

在 Nubint 中阅读本论文

本解读由 AI 根据论文原文生成,引用前请核对原文。