Abstract:Pan-sharpening aims to generate high-detail multi-spectral images (HRMS) through the fusion of panchromatic (PAN) and multi-spectral (MS) images. However, existing pan-sharpening methods often suffer from significant performance degradation when dealing with out-of-distribution data, as they assume the training and test datasets are independent and identically distributed. To overcome this challenge, we propose a novel frequency domain-irrelevant feature learning framework that exhibits exceptional generalization capabilities. Our approach involves parallel extraction and processing of domain-irrelevant information from the amplitude and phase components of the input images. Specifically, we design a frequency information separation module to extract the amplitude and phase components of the paired images. The learnable high-pass filter is then employed to eliminate domain-specific information from the amplitude spectrums. After that, we devised two specialized sub-networks (AFL-Net and PFL-Net) to perform targeted learning of the frequency domain-irrelevant information. This allows our method to effectively capture the complementary domain-irrelevant information contained in the amplitude and phase spectra of the images. Finally, the information fusion and restoration module dynamically adjusts the feature channel weights, enabling the network to output high-quality HRMS images. Through this frequency domain-irrelevant feature learning framework, our method balances generalization capability and network performance on the distribution of training dataset. Extensive experiments conducted on various satellite datasets demonstrate the effectiveness of our method for generalized pan-sharpening. Our proposed network outperforms state-of-the-art methods in terms of both quantitative metrics and visual quality, showcasing its superior ability to handle diverse, out-of-distribution data.

Learning High-frequency Feature Enhancement and Alignment for Pan-sharpening

SSETPAN: Spatial-Spectral Enhanced Transformer Based Network for Pansharpening

DuCoFPan: Dual-Condition Flow-based Network for Pan-sharpening

Adaptively Learning Low-high Frequency Information Integration for Pan-sharpening

Frequency Decoupled Domain-Irrelevant Feature Learning for Pan-sharpening

Pansharpening via Frequency-Aware Fusion Network With Explicit Similarity Constraints

Spatial and Spectral Extraction Network With Adaptive Feature Fusion for Pansharpening

HLF-Net: Pansharpening Based on High- and Low-Frequency Fusion Networks

Attention-Guided Progressive Frequency-Decoupled Network for Pan-Sharpening

Convolutional LSTM-Based Hierarchical Feature Fusion for Multispectral Pan-Sharpening

Pan-Sharpening With Wavelet-Enhanced High-Frequency Information

Frequency-Adaptive Pan-Sharpening with Mixture of Experts

PAN-Guided Band-Aware Multi-Spectral Feature Enhancement for Pan-Sharpening

Complementary Fusion Network Based on Frequency Hybrid Attention for Pansharpening

Pan-Sharpening Based on Multilevel Coupled Deep Network

Multiscale Dual-Domain Guidance Network for Pan-Sharpening

Domain-irrelevant Feature Learning for Generalizable Pan-sharpening

Modality-Aware Feature Integration for Pan-Sharpening

Progressive High-Frequency Reconstruction for Pan-Sharpening with Implicit Neural Representation

Hybrid Attention Based Residual Network for Pansharpening