NPSA: Nonorthogonal Principal Skewness Analysis.

Area of Science:

• Remote Sensing
• Signal Processing
• Data Analysis

Background:

• Principal Skewness Analysis (PSA) is a third-order generalization of Principal Component Analysis (PCA) for hyperspectral feature extraction.
• PSA transforms feature extraction into finding eigenpairs of a coskewness tensor, but orthogonal constraints can lead to deviations from actual eigenpairs.

Purpose of the Study:

• To address the limitations of PSA's orthogonal search strategy in hyperspectral imagery.
• To propose a new algorithm, Nonorthogonal Principal Skewness Analysis (NPSA), for more accurate eigenpair calculation and improved feature extraction.

Main Methods:

• Introduced a novel nonorthogonal search strategy for calculating eigenpairs of a coskewness tensor.
• Enlarged the eigenvector search space using the orthogonal complement of the Kronecker product of previous eigenvectors.
• Reduced algorithmic complexity through algebraic derivations.

Main Results:

• NPSA demonstrates the ability to obtain more accurate eigenpairs compared to PSA.
• The proposed nonorthogonal search strategy effectively overcomes the limitations of orthogonal constraints.
• Experiments on simulated and real hyperspectral data validate the effectiveness of NPSA for feature extraction.

Conclusions:

• NPSA offers a significant improvement over PSA for feature extraction in hyperspectral imagery.
• The method provides more accurate eigenpair solutions and enhanced feature extraction capabilities.
• NPSA presents a computationally efficient and valid approach for hyperspectral data analysis.