|
Proposing an Unsupervised Method for Extracting Reduced Spectral Features from Hyperspectral Images based on Band Clustering in Endmember Prototype Space
|
Ahmad Alizadeh Moghaddam *   , Mehdi Mokhtarzade |
| K. N. Toosi University of Technology |
|
|
Abstract: (444 Views) |
Hyperspectral images are a rich source of remote sensing data that are used in various studies, including agriculture, land cover/land use management and classification. However, the high dimensionality of feature space and limited number of training samples are among the main challenges in classifying these images and extracting information from them. Therefore, the use of feature reduction methods in two forms of feature selection and feature extraction is of great importance. Feature reduction methods are divided into two categories: supervised and unsupervised, that the unsupervised ones are more practical as they do not need any training data. In this study, an unsupervised method based on band clustering in endmember prototype space (EPBC) is presented. In this method, after estimating the virtual dimensionality of the image and extracting the endmembers, the endmember prototype space is formed and the bands are clustered using K-means clustering method in. Finally, the weighted mean of each cluster is extracted as a new feature. The final results obtained from classifying two hyperspectral images showed that the best overall accuracy of classification using the maximum likelihood classifier with features extracted by EPBC was %75.66 for the Indian Pines image and %89.71 for the Pavia University image, which outperformed well known unsupervised methods such as principal component analysis (PCA), minimum noise fraction (MNF), independent component analysis (ICA) and supervised method linear discriminant analysis (LDA). |
|
| Keywords: Hyperspectral Image, Prototype Space, Feature Extraction, Endmembers, Clustering, Classification |
|
|
Full-Text [PDF 1251 kb]
(95 Downloads)
|
Type of Study: Research |
Subject:
RS
Received: 2023/06/5 | Accepted: 2023/06/20 | ePublished ahead of print: 2024/09/28 | Published: 2024/10/29
|
|
|
|
|
|
|
| Send email to the article author |
|
|
