Deep Hyperspectral Image Sharpening.

Area of Science:

• Remote Sensing
• Computer Vision
• Image Processing

Background:

• Hyperspectral image (HSI) sharpening fuses low spatial resolution (LR) HSI with high spatial resolution (HR) multispectral images (MSI) to create HR-HSI.
• Existing HSI sharpening methods rely on image priors, which are sensitive to parameter selection and computationally intensive.

Purpose of the Study:

• To develop a novel deep learning-based HSI sharpening method (DHSIS) that directly learns image priors.
• To improve the accuracy and efficiency of HSI sharpening by integrating deep priors into a fusion framework.

Main Methods:

• A deep convolutional neural network-based residual learning approach is employed to learn image priors.
• The DHSIS method initializes HR-HSI by solving a Sylvester equation and then refines it using deep residual learning.
• Learned image priors are integrated back into the fusion framework for final HR-HSI reconstruction.

Main Results:

• Experimental results show DHSIS outperforms current state-of-the-art HSI sharpening techniques.
• The proposed method achieves superior reconstruction accuracy.
• DHSIS demonstrates a significant reduction in running time compared to existing approaches.

Conclusions:

• The DHSIS method offers an effective and efficient solution for hyperspectral image sharpening.
• Deep learning-based prior learning provides a robust alternative to traditional image prior modeling in HSI sharpening.