Change-Prior-Guided Unsupervised Change Detection of Heterogeneous Remote Sensing Images

Area of Science:

• Remote Sensing
• Geospatial Analysis
• Computer Vision

Background:

• Heterogeneous change detection (HeCD) uses multi-sensor remote sensing data to identify land-cover changes.
• Existing HeCD methods often focus on modality transformation and shared features, neglecting intrinsic change characteristics.
• This limitation hinders performance in complex change detection scenarios.

Purpose of the Study:

• To propose a novel unsupervised change prior-guided image transformation model (CPIT) for enhanced HeCD.
• To address the limitations of existing methods by incorporating intrinsic change characteristics.
• To improve the accuracy and effectiveness of land-cover change detection using heterogeneous remote sensing imagery.

Main Methods:

• Developed a change prior-guided image transformation model (CPIT) for unsupervised HeCD.
• Derived change semantic consistency and inconsistency rules from the nature of change detection.
• Encoded these rules as constraints, analogous to low-pass and high-pass spectral properties in graph signal processing.
• Integrated semantic constraints with sparsity and image transformation priors into a precise transformation model.

Main Results:

• The proposed CPIT model demonstrated improved performance in heterogeneous change detection.
• Experiments on five datasets validated the effectiveness of the CPIT approach.
• The method successfully identified land-cover changes by leveraging change priors.

Conclusions:

• The change prior-guided approach offers a more effective strategy for unsupervised HeCD.
• Incorporating intrinsic change characteristics and semantic rules enhances detection accuracy.
• The CPIT model provides valuable insights for future research in change detection.