Classification of Hyperspectral Imagery based on spectral gradient, SVM and spatial random forest

Abstract

Recently, many spectral-spatial-based methods have increased attention in hyperspectral image (HSI) classification. This paper proposes a novel method combining spectral gradient, SVM and spatial random forest (RF) for hyperspectral image classification, to better characterize the details and edges of the hyperspectral image. At the same time, it integrates the spectral and spatial features based on multiscale fusion. First, a spectral gradient technique is used to deal with the original hyperspectral data to acquire more intrinsic and comprehensive information. Then, the updated data is sent into the SVM to obtain probability output, and the spatial context information with different scales are further extracted. Finally, the multiscale spatial features are fused with the corresponding weights, and are subsequently fed as input into the random forest classifier. In experimental results, effectiveness of the proposed approach is confirmed by extensive experiments on three HSI datasets including AVIRIS Indian Pines, Salinas and ROSIS Pavia University. Compared with state-of-the-art methods, the proposed method obtains higher classification accuracy in terms of the overall accuracy and Kappa coefficient. Additionally, the proposed approach consumes lower running time compared with the comparison methods.