Windowed Linear Feature Extraction for Hyperspectral Image Processing

The rich spectral information provided by hyperspectral images (HSIs) are useful in discriminating between land cover classes. However, the limited number of training samples in comparison to the dimensionality of HSIs makes feature extraction a necessary step before classification. A common feature...

Full description

Saved in:
Bibliographic Details
Main Authors: Adebanjo, Hannah M., Tapamo, Jules R.
Format: Conference Proceeding
Language:English
Subjects:
component
Covariance matrices
Eigenvalues and eigenfunctions
Feature extraction
formatting
Hyperspectral imaging
insert
Iron
Microsoft Windows
Principal component analysis
style
styling
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The rich spectral information provided by hyperspectral images (HSIs) are useful in discriminating between land cover classes. However, the limited number of training samples in comparison to the dimensionality of HSIs makes feature extraction a necessary step before classification. A common feature extraction algorithm is the Principal Component Analysis (PCA) which has a drawback of discarding useful discriminative information in its lower eigenvectors. This paper proposes a new feature extraction method which addresses the limitation of PCA for HSI processing. The proposed approach which we refer to as Win-PCA attempts to reduce to reduce the loss of spatial information by partitioning the HSI into several windows and compute covariances for each window partition. Experiments performed to compare the conventional PCA and its other variants with WinPCA using the AVIRIS Indian Pine data show that the proposed WinPCA method performs better.
ISSN:2153-0033
DOI:10.1109/AFRICON46755.2019.9133907