Cycle-Based Frequency Disentanglement Diffusion Model With Self-Training for Cross-Domain Hyperspectral-RGB Change...

Area of Science:

• Remote Sensing
• Computer Vision
• Artificial Intelligence

Background:

• Hyperspectral image (HSI) change detection (CD) analyzes surface changes but is limited by data availability.
• Multimodal CD using HSI and RGB data addresses limitations but struggles with domain shifts.
• Existing domain adaptation methods face challenges with cross-domain multimodal CD due to modality differences.

Purpose of the Study:

• To develop a robust method for cross-domain HSI-RGB multimodal change detection.
• To enhance consistency in change representations across different modalities and domains.
• To overcome limitations of current domain adaptation techniques in multimodal CD.

Main Methods:

• A cycle-based frequency disentanglement diffusion model with self-training is proposed.
• A cyclic frequency domain disentanglement-based modality-domain alignment diffusion network achieves unified alignment.
• A curriculum-learning based self-training dual-domain CD network processes aligned images for collaborative CD.

Main Results:

• The proposed method significantly outperforms state-of-the-art approaches in cross-domain multimodal CD.
• The frequency-domain diffusion-driven self-training mechanism enhances consistency.
• Modality and domain alignment are achieved within a unified diffusion framework.

Conclusions:

• The developed model effectively addresses the challenges of cross-domain HSI-RGB multimodal CD.
• The approach demonstrates superior performance by leveraging frequency-domain diffusion and self-training.
• This work advances the field of multimodal change detection with improved domain adaptation capabilities.