Spectral-spatial classification of hyperspectral images using different spatial features and composite kernels

The high spectral and spatial resolution of hyperspectral images increases the capability to distinguish physical materials and objects, presenting new challenges to image analysis and classification. In fact, many studies have been conducted to extract and integrate spectral and contextual information in the classification process. However, the availability of various spatial features (e.g. morphological features, wavelet-based texture) and the inability to find the optimal one for all cases present a difficulty. In order to solve this problem, we proposed, in this paper, a novel spectral-spatial classification approach based on the integration of different spatial features. The proposed method exploited the performance of support vector machine (SVM) in the processing of data with high dimensionality and the properties of Mercer's kernels to construct composite kernels combining spectral and diverse spatial features. In this work, we employed all spectral information and two different spatial features which are Extended Multi-Attribute Profile (EMAP) and the mean of neighborhood pixels. The proposed kernels are compared to other multifeature SVM methods including stacked vector and traditional composite kernels. Experiments are conducted on the familiar AVIRIS "Indian Pines" data set. It was found that the proposed multifeature kernels provided more accurate classification results.