Papagbo/DMTS-HS-Unmixing
🌐 Unmix hyperspectral data using the DMTS-Net model, integrating a dual-stream architecture to enhance spectral variability analysis and model performance.
What's novel
🌐 Unmix hyperspectral data using the DMTS-Net model, integrating a dual-stream architecture to enhance spectral variability analysis and model performance.
Code Analysis
9 files read · 3 roundsA research codebase that performs blind hyperspectral unmixing on the Jasper Ridge dataset using a physics-based dispersion model (mass-spring oscillators) to generate spectral variability from VCA-extracted endmembers, combined with a 1D-CNN autoencoder for abundance estimation.
Strengths
The dispersion model provides a physically-motivated parametric approach to spectral variability that is more interpretable than purely data-driven methods. The training pipeline is well-structured with two-stage training, gradient clipping, early stopping, and learning rate scheduling.
Weaknesses
The EE Network's 4-layer FC backbone is completely dead code (param_layer is commented out), making the 'neural network' claim misleading — only the dispersion model's 4 scalar parameters per endmember are actually learned. The README describes a GUI application with installers, which is entirely inaccurate for this research codebase.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
13
Contributors
2
Files
24
Active weeks
4
Repository
Language
Python
Stars
1
Forks
0
License
—