[Orthogonal Vector Projection Algorithm for Spectral Unmixing].

Area of Science:

• Hyperspectral imaging
• Remote sensing
• Material science

Context:

• Spectrum unmixing is crucial for material quantity analysis in hyperspectral imagery.
• Traditional linear unmixing algorithms involve computationally intensive matrix operations (multiplication, inversion).
• These operations pose challenges for programming and hardware implementation, especially with increasing numbers of endmembers.

Purpose:

• To introduce Orthogonal Vector Projection, a novel method for hyperspectral spectrum unmixing.
• To simplify the unmixing process by avoiding matrix multiplication and inversion.
• To reduce computational complexity and enhance hardware applicability compared to existing algorithms.

Summary:

• Orthogonal Vector Projection computes orthogonal vectors via the Gram-Schmidt process for each endmember spectrum.
• Pixel signatures are projected onto these orthogonal vectors to directly obtain unconstrained abundances.
• This method avoids matrix inversion, relying on vector operations suitable for parallel computation and hardware.

Impact:

• Demonstrates lower computational complexity than Orthogonal Subspace Projection and Least Squares Error algorithms.
• Proves algorithm reasonability through its relationship with established methods.
• Validated through experimental results on synthetic and real hyperspectral images, confirming its effectiveness.