Multi-scale guided feature extraction and classification algorithm for hyperspectral images.

Area of Science:

• Remote Sensing
• Computer Vision
• Image Processing

Background:

• Traditional hyperspectral image classification struggles with object boundary distinction due to single-scale features, limiting accuracy.
• Object boundary details are crucial for accurate hyperspectral image classification.

Purpose of the Study:

• To propose a novel algorithm for hyperspectral image classification that effectively distinguishes object boundaries.
• To enhance classification accuracy by incorporating multi-scale spatial information.

Main Methods:

• Dimensionality reduction of hyperspectral data using Principal Component Analysis (PCA).
• Multi-scale spatial structure extraction via guided filtering with varying window sizes to preserve edge details.
• Classification using Support Vector Machines (SVM) with extracted multi-scale features.

Main Results:

• The proposed Multi-scale Guided Feature Extraction and Classification (MGFEC) algorithm extracts more accurate features than spectral information alone.
• Experimental results on diverse hyperspectral datasets demonstrate improved classification accuracy.
• The MGFEC algorithm shows superior performance compared to other spectral feature extraction methods.

Conclusions:

• The MGFEC algorithm effectively improves hyperspectral image classification accuracy by leveraging multi-scale spatial features.
• The proposed method is robust and suitable for processing various hyperspectral image datasets.