Spectral-Spatial Classification of Hyperspectral Images Based on Multifractal Features

Classification of hyperspectral images (HSIs) is an important step of HSI interpretation. Different studies demonstrate that spatial features can provide complementary information for increasing the accuracy of HSI classification. In this study, we propose a method of spectral-spatial classification of HSIs that is based on the use of specific multifractal features (MFs) as the spatial features. The proposed method of HSI classification consists of the following steps. First, informative MFs are extracted from first few principal components (PCs) of spectral features. For construction of the MFs, in windows centered on each element of PC images, using a multifractal image analysis, various local and global scaling exponents can be calculated. After that, obtained MFs are stacked with spectral features into high-dimensional feature vectors. Finally, the resulting high-dimensional vectors of spectral and multifractal features are classified by a support vector machine (SVM) or another classifier. Multifractal characteristics that are used to construct MFs have a lot of advantages: these characteristics provide a good textural separability of image objects, demonstrate an invariance to image scaling and rotation, and they are also insensitive to image noise. Experiments performed on two widely known test HSIs have demonstrated that proposed method exhibits better performance than competitive methods of spectral-spatial classification of HSIs, in terms of the overall accuracy (OA), average accuracy (AA) and kappa statistic. In addition, it is shown that the introduced classification method can outperform some deep learning methods of HSI classification, which in recent years have attracted great interest in HSI classification. Moreover, it was established that the proposed method can achieve good classification results if we use small training samples for classification.