Language-Guided Semantic Clustering for Remote Sensing Change Detection
Author Information
Author(s): Hu Shenglong, Bian Yiting, Chen Bin, Song Huihui, Zhang Kaihua, Vrochidis Stefanos
Primary Institution: Nanjing University of Information Science and Technology
Hypothesis
Can a language-guided semantic clustering framework improve remote sensing change detection by utilizing semantic embeddings?
Conclusion
The proposed LSC-CD framework achieves state-of-the-art performance in remote sensing change detection by effectively utilizing semantic information.
Supporting Evidence
- The LSC-CD framework utilizes a category text memory bank to enhance semantic modeling capabilities.
- Experimental results on three public benchmarks demonstrate the superiority of LSC-CD over existing methods.
- LSC-CD achieved an F1 score of 92.01 on the LEVIR-CD dataset, outperforming other methods.
Takeaway
This study shows that using words to help group similar changes in satellite images can make it easier to spot what has changed over time.
Methodology
The study employs a language-guided semantic clustering framework that integrates a CLIP model to enhance change detection in remote sensing images.
Limitations
The method relies on additional text semantic information and requires prior knowledge of categories.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website