Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

UniRSCD: A Unified Novel Architectural Paradigm for Remote Sensing Change Detection

Yuan Qu, Zhipeng Zhang, Chaojun Xu, Qiao Wan, Mengying Xie, Yuzeng Chen, Zhenqi Liu, Yanfei Zhong

TL;DR

UniRSCD tackles the fragmentation of change-detection tasks (BCD, SCD, BDA) by introducing a unified, single-stream architecture that fuses a state-space backbone with a Frequency Change Prompt Generator to capture both global context and fine-grained details. The approach eliminates the need for task-specific decoders by employing a simple, multi-task compatible decoder and prediction head, coupled with task-adaptive loss functions that balance accuracy and IoU across outputs. Empirical results on five public datasets show state-of-the-art or competitive performance for binary changes, semantic changes, and building-damage assessment, with notable efficiency gains from the unified design. Overall, UniRSCD demonstrates strong generalizability and practicality for rapid, robust change detection in diverse remote-sensing applications.

Abstract

In recent years, remote sensing change detection has garnered significant attention due to its critical role in resource monitoring and disaster assessment. Change detection tasks exist with different output granularities such as BCD, SCD, and BDA. However, existing methods require substantial expert knowledge to design specialized decoders that compensate for information loss during encoding across different tasks. This not only introduces uncertainty into the process of selecting optimal models for abrupt change scenarios (such as disaster outbreaks) but also limits the universality of these architectures. To address these challenges, this paper proposes a unified, general change detection framework named UniRSCD. Building upon a state space model backbone, we introduce a frequency change prompt generator as a unified encoder. The encoder dynamically scans bitemporal global context information while integrating high-frequency details with low-frequency holistic information, thereby eliminating the need for specialized decoders for feature compensation. Subsequently, the unified decoder and prediction head establish a shared representation space through hierarchical feature interaction and task-adaptive output mapping. This integrating various tasks such as binary change detection and semantic change detection into a unified architecture, thereby accommodating the differing output granularity requirements of distinct change detection tasks. Experimental results demonstrate that the proposed architecture can adapt to multiple change detection tasks and achieves leading performance on five datasets, including the binary change dataset LEVIR-CD, the semantic change dataset SECOND, and the building damage assessment dataset xBD.

UniRSCD: A Unified Novel Architectural Paradigm for Remote Sensing Change Detection

TL;DR

UniRSCD tackles the fragmentation of change-detection tasks (BCD, SCD, BDA) by introducing a unified, single-stream architecture that fuses a state-space backbone with a Frequency Change Prompt Generator to capture both global context and fine-grained details. The approach eliminates the need for task-specific decoders by employing a simple, multi-task compatible decoder and prediction head, coupled with task-adaptive loss functions that balance accuracy and IoU across outputs. Empirical results on five public datasets show state-of-the-art or competitive performance for binary changes, semantic changes, and building-damage assessment, with notable efficiency gains from the unified design. Overall, UniRSCD demonstrates strong generalizability and practicality for rapid, robust change detection in diverse remote-sensing applications.

Abstract

In recent years, remote sensing change detection has garnered significant attention due to its critical role in resource monitoring and disaster assessment. Change detection tasks exist with different output granularities such as BCD, SCD, and BDA. However, existing methods require substantial expert knowledge to design specialized decoders that compensate for information loss during encoding across different tasks. This not only introduces uncertainty into the process of selecting optimal models for abrupt change scenarios (such as disaster outbreaks) but also limits the universality of these architectures. To address these challenges, this paper proposes a unified, general change detection framework named UniRSCD. Building upon a state space model backbone, we introduce a frequency change prompt generator as a unified encoder. The encoder dynamically scans bitemporal global context information while integrating high-frequency details with low-frequency holistic information, thereby eliminating the need for specialized decoders for feature compensation. Subsequently, the unified decoder and prediction head establish a shared representation space through hierarchical feature interaction and task-adaptive output mapping. This integrating various tasks such as binary change detection and semantic change detection into a unified architecture, thereby accommodating the differing output granularity requirements of distinct change detection tasks. Experimental results demonstrate that the proposed architecture can adapt to multiple change detection tasks and achieves leading performance on five datasets, including the binary change dataset LEVIR-CD, the semantic change dataset SECOND, and the building damage assessment dataset xBD.

Paper Structure

This paper contains 14 sections, 8 equations, 3 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Deep Learning Change Detection
  4. State Space Models
  5. Method
  6. Overview
  7. Unified Change Frequency Prompt Encoder
  8. Unified Change Decoder and Prediction Head
  9. Multi-Task Adaptive Loss Function
  10. Experiments and Results
  11. Experimental Setup
  12. Comparison with State-of-the-Art Methods
  13. Diagnostic Study
  14. Conclusion

Figures (3)

  • Figure 1: Overall architectures for BCD task, SCD task, BDA task and UniRSCD. (a) BCD task obtains features representing changed objects. (b) SCD task obtains features representing semantic changes and binary changes, where the T1 and T2 decoders share the same prediction heads. (c) BDA task: Two features used for building localization and damage classification. (d) Our UniRSCD uses horizontal concatenation to process bitemporal images, then handles multiple tasks through a unified architecture.
  • Figure 2: The overall architecture of UniRSCD. The upper half of the figure illustrates the main workflow of UniRSCD. Horizontally concatenated bitemporal images are scanned along row and column directions and their inverses for input. Data flows sequentially to the right through processing units from Stage I to Stage IV, which include frequency change prompt generator (FCPG) and VSSBlock modules. After passing through the SAPF module and undergoing upsampling, the data is directed via UCPH to three parallel task branches: semantic change detection, binary change detection, and building change assessment. The lower half of the figure details the internal structure of the Frequency Change Prompt Generator (FCPG), Unified Change Prediction Head (UCPH), and Scale-Aware Prompt Fusion (SAPF).
  • Figure 3: Visualization of features at different layers of UniRSCD across three datasets.