Superpixel Tensor Sparse Coding for Structural Hyperspectral Image Classification

In this paper, a superpixel tensor sparse coding (STSC) based hyperspectral image classification (HIC) method is proposed, by exploring the high-order structure of hyperspectral image and utilizing information along all dimensions to better understand data. First, a hierarchical spatial affinity propagation algorithm is developed to rapidly cluster the image into multiple superpixels tensors. Then, a new STSC-based classifier followed by hybrid pixel-superpixel ensemble strategy is constructed for HIC. Because superpixels can reduce the misclassification caused by mixed pixel and tensor sparse coding can simultaneously classify multiple superpixels, rapid and accurate HIC can be achieved. Some experiments are taken on several datasets, and the results show the superiority of STSC to its counterparts in terms of speed and accuracy.