Manifold regularization for sparse unmixing of hyperspectral images

Area of Science:

• Remote Sensing
• Signal Processing
• Data Science

Background:

• Sparse unmixing is used for spectral mixture analysis in hyperspectral imaging.
• Traditional methods like collaborative sparse regression overlook hyperspectral data's geometric structure.

Purpose of the Study:

• To propose a novel sparse unmixing model for hyperspectral images.
• To enhance sparse regression by incorporating geometric data properties.

Main Methods:

• Developed a manifold regularized collaborative sparse regression model.
• Utilized graph Laplacian for manifold regularization to capture local geometry.
• Implemented an alternating direction method of multipliers algorithm for model optimization.

Main Results:

• The proposed manifold regularized model effectively integrates local geometric information.
• The developed algorithm efficiently solves the complex regression problem.

Conclusions:

• Experimental results confirm the model's superior performance on simulated and real hyperspectral data.
• The new approach advances sparse unmixing techniques in hyperspectral imaging.