Hyperspectral Image Classification Using Nonnegative Sparse Spectral Representation and Spatial Regularization
Hyperspectral image classification is an important technique for variety of applications and a challenge task because of high dimensional features and low SNR. This study proposes a robust hyperspectral image classification approach using nonnegative sparse spectral representation and spatial regularization. Due to the nonnegative composition of material spectra, we investigate a nonnegative sparse representation for spectral analysis, which can decompose the hyper-spectrum into the composite weight, as robust spectral feature, of the possible existed spectral prototypes (called as spectral dictionary) learned from the observed scene, and then conduct the pixel-wise recognition with the decomposed weights instead of raw spectrum. Furthermore, a spatial regularization method for category probability map propagation is explored for integrating the relationship among nearby pixels, and thus can provide more robust classification map. Experimental results on several hyperspectral image datasets validate that the proposed method can achieve much more accurate classification performance than the state-of-the-art approaches.
KeywordsHyperspectral image classification Nonnegative sparse representation Spatial regularization ADMM
- 9.Kuo, B.-C., Li, C.-H., Yang, J.-M.: Kernel nonparametric weighted feature extraction for hyperspectral image classification. IEEE Trans. Geosci. Remote Sens. 44(6), 1139–1155 (2009)Google Scholar
- 16.Samat, A., Du, P., Liu, S., Li, J., Cheng, L.: E2LMs: ensemble extreme learning machines for hyperspectral image classification. IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens. 7(4), 1060–1069 (2014)Google Scholar
- 17.Hu, W., Huang, Y., Wei, L., Zhang, F., Li, H.: Deep convolutional neural networks for hyperspectral image classification. J. Sens. 2015, Article no. 258619 (2015)Google Scholar