Spatial-Spectral Constrained Adaptive Graph for Hyperspectral Image Clustering.

Area of Science:

• Remote Sensing
• Computer Vision
• Data Science

Background:

• Hyperspectral image (HSI) clustering is vital for pixel classification but is hindered by high dimensionality and noise.
• Existing methods struggle to effectively capture local geometric structures and mitigate noise interference.

Purpose of the Study:

• To propose a novel Projected Clustering with Spatial-Spectral Constrained Adaptive Graph (PCSSCAG) method for robust HSI clustering.
• To address the challenges of high dimensionality and noise corruption in hyperspectral data.

Main Methods:

• PCSSCAG constructs an adaptive adjacency graph to capture local data structures.
• It employs spatial-spectral constraints to leverage both spatial and spectral information, reducing noise impact.
• Projection learning is integrated to minimize redundancy and computational cost.

Main Results:

• Experimental results on diverse HSI datasets demonstrate the effectiveness of PCSSCAG.
• The proposed method shows superior performance compared to existing HSI clustering techniques.
• PCSSCAG successfully reduces the negative influence of noise on graph construction.

Conclusions:

• The PCSSCAG method offers a superior approach to hyperspectral image clustering.
• It effectively balances spatial-spectral information exploration and noise reduction.
• The method provides a computationally efficient and accurate solution for HSI clustering tasks.