Improving the performance of RPNet with LDA for extracting the deep features for the classification of hyperspectral images
|
Behnam, Asghari Beirami * , Mehdi Mokhtarzade |
K. N .Toosi university |
|
Abstract: (2593 Views) |
In recent years, deep models have achieved great success in various fields of image processing. These models have been used in some research fields of hyperspectral data processing, such as; classification and target detection. The random patches network (RPNet) has recently been proposed to extract hierarchical deep features for hyperspectral image classification. RPNet is important as it is an unsupervised method, and as a consequence, it has a fast performance to extract deep features. Despite the good performance of this network, due to the usage of the principal component analysis (PCA) method in its main structure, maximum discrimination between classes is not guaranteed in extracted features. Therefore, in this paper,in order to improve the performance of RPNet, a new method called LDA-RPNet based on linear discriminant analysis (LDA) is proposed. Experiments on two hyperspectral datasets, Indian Pines and Pavia University, show that the LDA-RPNet can extract more compact and suitable features for classifying hyperspectral images. Also, based on the experiments, the LDA-RPNet can increase the overall accuracy by up to 2.5% compared to the classical RPNet. |
|
Keywords: Hyperspectral image classification, Random patches network, linear discriminant analysis, Principal component analysis, Hierarchical deep features. |
|
Full-Text [PDF 1658 kb]
(748 Downloads)
|
Type of Study: Research |
Subject:
RS Received: 2019/11/16 | Accepted: 2020/05/9 | ePublished ahead of print: 2022/04/12 | Published: 2022/06/8
|
|
|
|
|
Send email to the article author |
|