Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Disentangling Hardness from Noise: An Uncertainty-Driven Model-Agnostic Framework for Long-Tailed Remote Sensing Classification

Chi Ding, Junxiao Xue, Xinyi Yin, Shi Chen, Yunyun Shi, Yiduo Wang, Fengjian Xue, Xuecheng Wu

TL;DR

This work tackles long-tailed remote sensing classification by disentangling hardness from noise through an uncertainty-driven, model-agnostic framework called DUAL. Grounded in Evidential Deep Learning, DUAL decomposes predictive uncertainty into Epistemic Uncertainty (EU) and Aleatoric Uncertainty (AU), using EU to reweight hard tail samples and AU to adaptively smooth labels for noisy data, all within a Dirichlet-distribution-based training scheme. Empirical results on DOTA, DIOR, and FGSC-23 across multiple backbones show consistent tail-class and overall accuracy gains, outperforming baselines like TGN and SADE and supported by ablations. The approach offers a practical, generalizable path to robust long-tailed remote sensing classification with clear uncertainty-based training signals.

Abstract

Long-Tailed distributions are pervasive in remote sensing due to the inherently imbalanced occurrence of grounded objects. However, a critical challenge remains largely overlooked, i.e., disentangling hard tail data samples from noisy ambiguous ones. Conventional methods often indiscriminately emphasize all low-confidence samples, leading to overfitting on noisy data. To bridge this gap, building upon Evidential Deep Learning, we propose a model-agnostic uncertainty-aware framework termed DUAL, which dynamically disentangles prediction uncertainty into Epistemic Uncertainty (EU) and Aleatoric Uncertainty (AU). Specifically, we introduce EU as an indicator of sample scarcity to guide a reweighting strategy for hard-to-learn tail samples, while leveraging AU to quantify data ambiguity, employing an adaptive label smoothing mechanism to suppress the impact of noise. Extensive experiments on multiple datasets across various backbones demonstrate the effectiveness and generalization of our framework, surpassing strong baselines such as TGN and SADE. Ablation studies provide further insights into the crucial choices of our design.

Disentangling Hardness from Noise: An Uncertainty-Driven Model-Agnostic Framework for Long-Tailed Remote Sensing Classification

TL;DR

This work tackles long-tailed remote sensing classification by disentangling hardness from noise through an uncertainty-driven, model-agnostic framework called DUAL. Grounded in Evidential Deep Learning, DUAL decomposes predictive uncertainty into Epistemic Uncertainty (EU) and Aleatoric Uncertainty (AU), using EU to reweight hard tail samples and AU to adaptively smooth labels for noisy data, all within a Dirichlet-distribution-based training scheme. Empirical results on DOTA, DIOR, and FGSC-23 across multiple backbones show consistent tail-class and overall accuracy gains, outperforming baselines like TGN and SADE and supported by ablations. The approach offers a practical, generalizable path to robust long-tailed remote sensing classification with clear uncertainty-based training signals.

Abstract

Long-Tailed distributions are pervasive in remote sensing due to the inherently imbalanced occurrence of grounded objects. However, a critical challenge remains largely overlooked, i.e., disentangling hard tail data samples from noisy ambiguous ones. Conventional methods often indiscriminately emphasize all low-confidence samples, leading to overfitting on noisy data. To bridge this gap, building upon Evidential Deep Learning, we propose a model-agnostic uncertainty-aware framework termed DUAL, which dynamically disentangles prediction uncertainty into Epistemic Uncertainty (EU) and Aleatoric Uncertainty (AU). Specifically, we introduce EU as an indicator of sample scarcity to guide a reweighting strategy for hard-to-learn tail samples, while leveraging AU to quantify data ambiguity, employing an adaptive label smoothing mechanism to suppress the impact of noise. Extensive experiments on multiple datasets across various backbones demonstrate the effectiveness and generalization of our framework, surpassing strong baselines such as TGN and SADE. Ablation studies provide further insights into the crucial choices of our design.
Paper Structure (15 sections, 17 equations, 2 figures, 6 tables)

This paper contains 15 sections, 17 equations, 2 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Long-tailed Learning
  4. Uncertainty Estimation
  5. Methodology
  6. Evidential Deep Learning
  7. Epistemic Uncertainty and Aleatoric Uncertainty
  8. Uncertainty-aware long-tailed learning
  9. Experiments
  10. Main Results
  11. Relationship of Different Epistemic Uncertainty
  12. Backbone Generalization
  13. Hyperparameter Analysis
  14. Ablation Study
  15. Conclusion and Discussions

Figures (2)

  • Figure 1: The overview of our proposed DUAL framework. The pipeline consists of three stages: (1) Evidential Deep Learning, which predicts class-level evidence from the backbone; (2) Uncertainty Decomposition, which decomposes prediction uncertainty into EU and AU; and (3) Uncertainty-aware Learning, where EU serves as an indicator of sample scarcity to reweight hard tail samples, while AU quantifies data ambiguity to guide adaptive label smoothing for noise suppression.
  • Figure 2: Correlation between the proposed K/S metric and entropy-based EU on FGSC-23 dataset.