Semi-Supervised Hyperspectral Unmixing via the Weighted Lasso

In this paper a novel approach for semi-supervised hyperspectral unmixing is presented. First, it is shown that this problem inherently accepts a sparse solution. Then, based on this observation, an efficient ℓ1 regularized least squares algorithm is proposed, in which the constraints that are naturally imposed to the problem, are suitably incorporated. Simulations results show that the proposed method achieves the performance of quadratic programming based techniques with much lower computational requirements.

• Type: Conference
• Year: 2010
• Where: ICASSP
• Authors: Konstantinos Themelis, Athanasios A. Rontogiannis, Konstantinos Koutroumbas
• Keywords: ICASSP 2010, Lower Computational Requirements, Semi-supervised Hyperspectral, Signal Processing, Sparse Solution