Supervised Polarimetric SAR Image Classification Using Tensor Local Discriminant Embedding.

Area of Science:

• Remote Sensing
• Computer Vision
• Signal Processing

Background:

• Polarimetric Synthetic Aperture Radar (PolSAR) image classification is crucial for Earth observation.
• Existing dimensionality reduction (DR) methods often neglect spatial context, limiting classification performance.
• Speckle noise in PolSAR data further degrades classification accuracy.

Purpose of the Study:

• To develop a novel feature extraction method for supervised PolSAR image classification.
• To address the limitations of traditional DR methods by incorporating spatial information.
• To improve the robustness and accuracy of PolSAR image classification.

Main Methods:

• A Tensor Local Discriminant Embedding (TLDE) method is proposed for feature extraction.
• Each pixel is represented as a third-order tensor, capturing spatial (patch) and polarimetric information.
• Supervised dimensionality reduction is applied to project tensors into a low-dimensional feature space.

Main Results:

• The TLDE method significantly enhances classification accuracy on real and simulated PolSAR datasets.
• The proposed approach effectively mitigates the impact of speckle noise on classification.
• Classification performance is validated using Nearest Neighbor (NN) and Support Vector Machine (SVM) classifiers.

Conclusions:

• TLDE offers a superior approach to feature extraction for PolSAR image classification.
• Integrating spatial and polarimetric information via tensor representation is key to improved performance.
• The method provides a robust solution for accurate and noise-resilient PolSAR image analysis.