Abstract
Remote sensing change detection (CD) using multitemporal hyperspectral images (HSI) is a process of extraction of change features and classification. However, the high dimensionality of HSI not only leads to expensive computation but also suffers from spectral-spatial variability and inner-class heterogeneity. In this paper, we proposed two algorithms for CD based on the tensor train (TT) decomposition, which uses a well-balanced matricization strategy to capture hidden information from tensors. The first algorithm TT decomposition uses nuclear norm hence named TTNN_CD and the second algorithm uses multilinear matrix factorization bypassing the expensive SVD named TTMMF_CD. We use -augmentation (KA) scheme to represent the low-order tensor into a high-order tensor to extract change features efficiently. The experiments reveal that TT-based CD outperforms its tensor counterpart, HOSVD, and some other commonly used approaches.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have