.png){kind=icon}
.png){kind=icon}
Return
Nesterov-accelerated non-negative matrix factorization unrolling network for hyperspectral unmixing
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2025.131202.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Deep learning is making notable progress in hyperspectral unmixing (HU), with algorithm unrolling networks providing advantages in interpretability compared to traditional networks. These networks mainly focused on leveraging prior knowledge from conventional algorithms to enhance their performance. However, research on improving network learning efficiency remains limited, specifically in terms of accuracy of endmember and abundance estimation, as well as the number of iterations and training time. In this paper, we propose a Nesterov-accelerated non-negative matrix factorization unrolling network (NANMF-Net) for HU, which not only improves unmixing accuracy but also reduces the number of training iterations. Specifically, NANMF-Net adopts a two-stage network structure. The first stage uses an unrolling NMF structure, which focuses on learning relevant information. The second stage applies the Nesterov’s method to optimize the network, reinforcing the connections among layers. These two stages work together to achieve more accurate endmember and abundance estimates with fewer training iterations. NANMF-Net demonstrated superior unmixing performance and fewer iterations on both synthetic and real datasets compared to some traditional models and networks.
Keywords:
deep learning
hyperspectral unmixing
non-negative matrix factorization
unrolling network
Nesterov acceleration
Journal
IF:
6.5
Papers:
2.5W
Citations:
6.5W