Behnam Asghari Beirami; Mehdi Mokhtarzade
Abstract
The use of spatial features to improve the classification accuracy of hyperspectral images has become popular in recent years. Various methods for spectral-spatial classification of hyperspectral images have been introduced to date, and relevant research is being conducted to develop methods with a more ...
Read More
The use of spatial features to improve the classification accuracy of hyperspectral images has become popular in recent years. Various methods for spectral-spatial classification of hyperspectral images have been introduced to date, and relevant research is being conducted to develop methods with a more straightforward structure and higher accuracy. This paper introduces a new method for producing efficient features for classifying hyperspectral images based on combining extracted features from the weighted local kernel matrix of spectral and fractal features. One of the main advantages of weighted local kernel matrices is that they model nonlinear dependencies between features that are not taken into account by traditional feature generation methods. In this study's proposed method, the weighted kernel local matrix method is used in order to generate new features from spectral features and directional fractal dimension features. Then these two feature vectors are joined together for each pixel and form a vector rich in spectral-spatial information. Finally, to determine each pixel's label, the obtained feature vector is classified by the support vector machine (SVM) algorithm. The results obtained from two real hyperspectral images of Indian Pine and Pavia University show that the accuracies in the proposed method are above 98% on average in both data sets, which is more than 5% higher than the average accuracy of several other hyperspectral image classification methods.
