Joint and Progressive Subspace Analysis (JPSA) With Spatial-Spectral Manifold Alignment for Semisupervised...

Area of Science:

• Remote Sensing
• Computer Vision
• Machine Learning

Background:

• Conventional nonlinear subspace learning methods for hyperspectral data face challenges in explainability, cost-effectiveness, generalization, and representability.
• Existing techniques often struggle with explicit mapping, linearization, out-of-sample data, and spatial-spectral discrimination.

Purpose of the Study:

• To develop a novel linearized subspace analysis technique for semisupervised hyperspectral dimensionality reduction (HDR).
• To overcome the limitations of traditional methods by introducing spatial-spectral manifold alignment.
• To achieve a high-level, semantically meaningful, joint spatial-spectral feature representation for hyperspectral data.

Main Methods:

• Joint and Progressive Subspace Analysis (JPSA) is proposed, integrating latent subspace learning with a linear classifier.
• The method employs a progressive search through intermediate subspace states for optimal mapping.
• Spatial and spectral manifold alignment is utilized to preserve topological properties between original and compressed data.

Main Results:

• JPSA demonstrates superior performance in hyperspectral dimensionality reduction compared to state-of-the-art methods.
• Experiments on the Indian Pines dataset achieved 92.98% accuracy, and on the University of Houston dataset, 86.09% accuracy.
• The effectiveness was validated using a nearest neighbor (NN) classifier.

Conclusions:

• JPSA offers an effective solution for semisupervised hyperspectral dimensionality reduction.
• The technique successfully addresses limitations in explainability, generalization, and representability.
• The proposed method significantly improves classification accuracy on hyperspectral datasets.