Heterospectral Structure Compensation Sampling for Hyperspectral Fusion Computational Imaging.

Area of Science:

• Computational Imaging
• Remote Sensing
• Image Processing

Background:

• Current hyperspectral fusion methods use high-resolution multispectral images (HRMSI) for spatial details in low-resolution hyperspectral images (LRHSI).
• These methods are limited by HRMSI's low spectral resolution, hindering effective information transfer for finer spectral ranges in LRHSI.

Purpose of the Study:

• To enhance hyperspectral computational imaging accuracy by addressing limitations in existing fusion techniques.
• To introduce novel mechanisms for improved spatial-temporal information fusion in hyperspectral imaging.

Main Methods:

• Proposed Heterospectral Structure Compensation Sampling (HSC-sampling) mechanism to analyze structural complementarity among LRHSI bands, compensating for missing details.
• Developed a Multi-phase Mixed Modeling (M2M) approach to extract fusion features from three phases, creating a multi-variate mixed cube for high-dimensional HSI data.
• Constructed a Merging Residual Concatenation (MRC) network integrating HSC-sampling and M2M for hyperspectral fusion computational imaging.

Main Results:

• The proposed MRC network significantly outperforms state-of-the-art methods in hyperspectral fusion performance across multiple datasets.
• Demonstrated the effectiveness of the HSC-sampling mechanism in various hyperspectral imaging tasks.
• Achieved improved accuracy in reconstructing hyperspectral images by leveraging spectral and spatial information more effectively.

Conclusions:

• The HSC-sampling mechanism and M2M approach offer a significant advancement in hyperspectral fusion computational imaging.
• The developed MRC network provides a robust solution for accurate hyperspectral image reconstruction.
• The findings pave the way for more sophisticated hyperspectral imaging applications requiring high spectral and spatial resolution.