Wavelet-Guided Dual-Frequency Encoding for Remote Sensing Change Detection

2025-08-07

AI Summary

In remote sensing change detection, existing spatial-domain methods struggle to model subtle changes—particularly at object boundaries—due to limited feature diversity and discriminability. To address this, we propose a waveguide dual-frequency encoding framework, the first to integrate wavelet transform into change detection. Our approach employs discrete wavelet decomposition to decouple high-frequency components (edges and fine details) from low-frequency components (global structures), enabling explicit frequency-domain modeling. We design a frequency-domain difference enhancement module and a progressive contextual differencing module, jointly optimized within a Transformer architecture to enhance both local sensitivity and global discriminability. Evaluated on multiple benchmark datasets, our method significantly mitigates edge blurring and achieves state-of-the-art performance in mF1 and IoU metrics. Results demonstrate that frequency-domain modeling substantially improves the detection of subtle changes, offering enhanced effectiveness and robustness.

Technology Category

• Computer Vision: Remote Sensing / Geospatial AI
• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Search and Retrieval-Augmented AI: Vertical and domain-specific search

Abstract

Change detection in remote sensing imagery plays a vital role in various engineering applications, such as natural disaster monitoring, urban expansion tracking, and infrastructure management. Despite the remarkable progress of deep learning in recent years, most existing methods still rely on spatial-domain modeling, where the limited diversity of feature representations hinders the detection of subtle change regions. We observe that frequency-domain feature modeling particularly in the wavelet domain an amplify fine-grained differences in frequency components, enhancing the perception of edge changes that are challenging to capture in the spatial domain. Thus, we propose a method called Wavelet-Guided Dual-Frequency Encoding (WGDF). Specifically, we first apply Discrete Wavelet Transform (DWT) to decompose the input images into high-frequency and low-frequency components, which are used to model local details and global structures, respectively. In the high-frequency branch, we design a Dual-Frequency Feature Enhancement (DFFE) module to strengthen edge detail representation and introduce a Frequency-Domain Interactive Difference (FDID) module to enhance the modeling of fine-grained changes. In the low-frequency branch, we exploit Transformers to capture global semantic relationships and employ a Progressive Contextual Difference Module (PCDM) to progressively refine change regions, enabling precise structural semantic characterization. Finally, the high- and low-frequency features are synergistically fused to unify local sensitivity with global discriminability. Extensive experiments on multiple remote sensing datasets demonstrate that WGDF significantly alleviates edge ambiguity and achieves superior detection accuracy and robustness compared to state-of-the-art methods. The code will be available at https://github.com/boshizhang123/WGDF.

Problem

• Enhances detection of subtle changes in remote sensing images
• Improves edge detail representation via frequency-domain modeling
• Combines local and global features for accurate change detection

Innovation

• Wavelet-Guided Dual-Frequency Encoding (WGDF)
• Dual-Frequency Feature Enhancement (DFFE) module
• Frequency-Domain Interactive Difference (FDID) module