Abstract:In remote sensing image change detection (CD), leveraging distant contextual information is crucial for accuracy. Traditional neural network training, which involves image cropping, limits the perception of such information in large-scale images. To overcome this, we propose the "Panoramic Change Analysis Framework (PCA-F)" for high-resolution remote sensing CD. This innovative framework features a dual-branch architecture to integrate broad area information and uses advanced spatial alignment for merging wide-area and local features. PCA-F overcomes traditional method limitations and integrates well with various CD networks. The experimental results demonstrate a significant improvement in accuracy for the three baseline methods on two datasets (BIT: 3.51%, 4.66%; Changeformer: 1.49%, 2.64%; RDPNet: 1.97%, 5.31%), offering a new direction for enhancing CD performance in remote sensing imagery.

Panoramic Change Analysis Framework (PCA-F): A New Method for Large-Scale Change Detection in High-Resolution Remote Sensing Images