Improvement of remote sensing classification method by multiway support tensor machine

In remote sensing image classification, it is usually to introduce a spectral feature vector by transferring the digital number of a pixel from each band into an array. However, this kind of vector represents only one pixel of a remote sensing image, considers the spectral information but ignores the spatial relationship of neighboring pixels. In this paper, we propose a multiway support tensor machine for remote sensing image classification. The training samples are represented as 3-order tensors with local neighbor information, then, the mathematical model and solution of multiway support tensor machine are discussed in detail. Experiments on the classification of HYDICE hyperspectral data set suggest that this scheme can deliver a high classification rate with a small number of training samples.