Iterative Pseudo Label Generation for Hyperspectral Image Classification
Author Information
Author(s): Zhao Zheng, Zhou Guangyao, Wang Qixiong, Feng Jiaqi, Jiang Hongxiang, Zhang Guangyun, Zhang Yu
Primary Institution: Beihang University
Hypothesis
Can an Iterative Pseudo Label Generation framework improve semi-supervised hyperspectral image classification using the Segment Anything Model?
Conclusion
The proposed framework significantly enhances classification performance in hyperspectral image analysis, even with limited annotations.
Supporting Evidence
- Experiments on the Indian Pines and Pavia University datasets show significant performance improvements over existing methods.
- The framework effectively utilizes both labeled and pseudo-labeled data to enhance model training.
- Iterative refinement of pseudo labels leads to higher classification accuracy.
Takeaway
This study shows a new way to help computers understand images taken with special cameras by using smart guessing to fill in the gaps when there aren't enough labeled examples.
Methodology
The study uses an Iterative Pseudo Label Generation framework that combines a small number of labeled data with a large amount of unlabeled data to improve classification accuracy.
Limitations
The performance can be affected by the method used to group spectral bands and the quality of initial annotated labels.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website