Hyperspectral Image Few-shot Classification Based on Analogous Tensor Decomposition

The main idea of few-shot learning is to learn a generalized model which can realize the transfer of knowledge from the source domain to the target domain with a small number of labeled samples. However, existing few-shot learning algorithms for hyperspectral image (HSI) simply and directly extract the spatial-spectral features from the three-dimensional local neighborhood of HSI pixels. In other words, it lacks an effective mathematical models for analyzing and preprocessing the local spatial and spectral information of HSI pixels. Therefore, a HSI few-shot classification algorithm based on analogous tensor decomposition (HSIFC-ATD) is proposed in this paper. By applying the analogous tensor decomposition to the 3-D local spatial-spectral tensor of HSI pixels, redundant information contained in the 3-D local spatial-spectral tensor is effectively removed, which improves the efficiency of subsequent HSI few-shot classification. The experimental results show that the proposed HSIFC-ATD algorithm achieves excellent classification performance on HSI few-shot classification.