一种结合记忆与深度展开的高光谱快照压缩成像重建方法
By combining memory and deep unfolding methods, introducing a degradation learning module and a memory-enhanced spatial-spectral attention mechanism, the robustness and stability issues of the inverse problem in hyperspectral snapshot compressed imaging reconstruction are solved, achieving high-quality image reconstruction results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHONGQING UNIV
- Filing Date
- 2025-08-08
- Publication Date
- 2026-07-17
AI Technical Summary
Existing hyperspectral snapshot compression imaging reconstruction methods are highly ill-posed to the inverse problem and sensitive to noise, making it difficult to achieve efficient and robust reconstruction. Furthermore, deep learning methods lack physical interpretability and stability, failing to fully exploit the spatial-spectral features of hyperspectral data.
A hyperspectral snapshot compression imaging reconstruction method combining memory and deep unfolding is proposed. By introducing a degradation learning module and a memory-enhanced spatial-spectral attention mechanism, the optimal solution is gradually approximated. The interpretability of the deep unfolding inference framework and physical model is utilized to dynamically correct the linear projection process, enhance the inter-stage feature transferability, and capture the spatial details and spectral features of the image.
It improves the accuracy and stability of hyperspectral image reconstruction, enhances reconstruction quality and system robustness, better recovers image details and global dependencies, and improves the model's feature utilization efficiency and generalization ability.
Smart Images
Figure CN120976434B_ABST