Related Experiment Videos
Adaptive Heterogeneity-Aware Tensor Decomposition for Hyperspectral Image Denoising
1School of Information and Communication Engineering, Beijing University of Posts and Telecommunications, Beijing 100876, China.
Sensors (Basel, Switzerland)
|July 28, 2026
Summary
Adaptive Heterogeneity-Aware Tensor Decomposition (AHTD) enhances hyperspectral image denoising by selectively refining heterogeneous regions. This approach improves denoising performance and efficiency compared to global methods.
Area of Science:
- Remote Sensing
- Computer Vision
- Signal Processing
Background:
- Hyperspectral image denoising requires balancing spectral consistency with spatial variations.
- Existing tensor decomposition methods often struggle with spatially heterogeneous data.
Purpose of the Study:
- To introduce Adaptive Heterogeneity-Aware Tensor Decomposition (AHTD) for improved hyperspectral image denoising.
- To develop a computationally efficient refinement module for tensor-based hyperspectral processing.
Main Methods:
- AHTD augments global Tucker decomposition with heterogeneity-guided local tensor shrinkage.
- It estimates spectral subspaces, detects heterogeneous regions using variance and edge analysis.
- Adaptive singular-value shrinkage is applied, modulated by regional patch complexity.
Main Results:
- AHTD achieved 0.35-0.40 dB PSNR gains over the global Tucker baseline.
- Maintained or improved SSIM, SAM, and ERGAS metrics.
- Offered threefold runtime reduction compared to dense local refinement.
Conclusions:
- AHTD provides a computationally efficient and interpretable refinement for hyperspectral tensor denoising.
- Selective refinement based on heterogeneity significantly enhances denoising performance.
- The method effectively reconciles global spectral coherence with local spatial details.