We consider the tensor-based spectral-spatial featureudextraction problem for hyperspectral image classification.udFirst, a tensor framework based on circular convolution is proposed.udBased on this framework, we extend the traditional PCA toudits tensorial version TPCA, which is applied to the spectral-spatialudfeatures of hyperspectral image data. The experiments showudthat the classification accuracy obtained using TPCA featuresudis significantly higher than the accuracies obtained by its rivals.
展开▼
