ePrints@IIScePrints@IISc Home | About | Browse | Latest Additions | Advanced Search | Contact | Help

Four-directional spatial regularization for sparse hyperspectral unmixing

Ahmad, T and Lyngdoh, Rb and Sahadevan, As and Raha, S and Gupta, PK and Misra, A (2020) Four-directional spatial regularization for sparse hyperspectral unmixing. In: Journal of Applied Remote Sensing, 14 (4). 046511.

[img] PDF
jou_app_rem_14-4_2020.pdf - Published Version
Restricted to Registered users only

Download (13MB) | Request a copy
Official URL: https://doi.org/10.1117/1.JRS.14.046511


A four-directional total variation technique is proposed to encapsulate the spatial contextual information for sparse hyperspectral image (HSI) unmixing. Traditional sparse total variation techniques explore gradient information along with the horizontal and vertical directions. As a result, spatial disparity due to high noise levels within the neighboring pixels are not considered while unmixing. Moreover, oversmoothing due to total variation may depreciate the spatial details in the abundance map. In this context, we propose a four-directional regularization technique (Sparse Unmixing with Splitting Augmented Lagrangian: Four-Directional Total Variation, SUnSAL-4DTV) for sparse unmixing. The four-directional total variation scheme is transformed into the fast-Fourier-transform domain to reduce the higher computational requirements. An alternating-direction-method-of-multipliers-based iterative scheme is proposed for solving the large-scale optimization problem. An adaptive scheme is introduced to update the regularization parameters to ensure faster convergence. Extensive numerical simulations were conducted on both simulated and real hyperspectral datasets to demonstrate the robustness of proposed technique. Comparative analysis on noisy (low signal-to-noise-ratio) HSIs shows the robustness of SUnSAL-4DTV over the state-of-the-art algorithms.

Item Type: Journal Article
Publication: Journal of Applied Remote Sensing
Publisher: SPIE
Additional Information: The copyright of the article belongs to the SPIE .
Keywords: alternating direction multipliers method; hyperspectral image; sparse unmixing; total variation spatial regularization
Department/Centre: Division of Interdisciplinary Sciences > Computational and Data Sciences
Date Deposited: 03 Sep 2022 06:05
Last Modified: 03 Sep 2022 06:05
URI: https://eprints.iisc.ac.in/id/eprint/76415

Actions (login required)

View Item View Item