|
The Classification of Hyperspectral Images Using the Weighted Local Kernel Matrix Network and Support Vector Machines Classifier
|
Behnam Asghari Beirami *   , Mehdi Mokhtarzade |
| K. N. Toosi University of Technology |
|
|
Abstract: (476 Views) |
Over the last few years, deep learning models have received a lot of attention for the spectral-spatial classification of the hyperspectral images. One of the significant advantages of deep learning methods is that they incorporate both spatial and spectral information in classifying hyperspectral images. Although these models produce accurate classified maps, they are computationally complex, and precise settings of their parameters require a large number of training samples. In order to address these issues, a simplified method that can efficiently extract the spectral-spatial information from a hyperspectral image must be developed. Therefore, the current study proposed a new method for generating the spectral-spatial features of the hyperspectral images. The proposed method uses weighted local kernel matrix representation and minimum noise fraction transformation sequentially and repetitively in order to generate deep spectral-spatial features. The proposed network's spectral-spatial features, which show the local nonlinear relationship between the features extracted from the components of the minimum noise fraction transform at different depths, will finally be stacked together and fed into the support vector machine algorithm for classification. Two hyperspectral benchmark images of the Indian Pines and data from Pavia University are used to test the proposed algorithm. The performance of the proposed method is compared to the spectral classification method and four other spectral-spatial classification methods proposed in recent years. Comparisons show that the proposed method is more accurate in the Indian Pines image more than 20% and in Pavia university image more than 10% than the image classification using the spectral features. In addition, the proposed method is 1% more accurate than the other four spectral-spatial classification methods on average.
|
|
| Keywords: Weighted local kernel matrix representation, Classification, Hyperspectral, Feature extraction |
|
|
Full-Text [PDF 1125 kb]
(139 Downloads)
|
Type of Study: Research |
Subject:
RS
Received: 2021/10/20 | Accepted: 2022/05/1 | Published: 2024/06/20
|
|
|
|
|
|
|
| Send email to the article author |
|
|
