Skip to Main content Skip to Navigation
Preprints, Working Papers, ...

Multifeature Hyperspectral Unmixing Based on Tensor Decomposition

Abstract : Hyperspectral unmixing is an active area of blind source separation. It refers to the representation of mixed pixels (samples) as a set of pure materials (sources), weighted by their abundances. Since spectral features alone are often insufficient, it is common to rely on other features of the scene as additional knowledge. In this paper, the hyperspectral image (HSI) is represented as a high-order tensor with additional features in a multimodal, multifeature framework, spanning modes of pixels, spectral features, and additional features, where matrix models become insufficient. This requires the use of tensor models, and particularly the Canonical Polyadic Decomposition, which is blind and straightforward for unmixing, and maintains the physical properties of the data. So far, this model has been applied in preliminary and specific applications, and still lacks a general framework for unmixing including the interpretation of the results. In this paper, we propose a methodological framework for multifeature unmixing based on the Alternating Optimization Alternating Direction Method of Multipliers algorithm and incorporating Abundance Sum-to-one Constraint (AO-ADMM-ASC), with in-depth mathematical, physical and graphical interpretations and links to the Extended Linear Mixing Model. Moreover, we propose to incorporate Mathematical Morphology as spatial features in multifeature unmixing and revise the work of patch features in order to demonstrate the interest of the proposed framework. Experiments on real HSI data sets show the efficiency of AO-ADMM-ASC and allows an in-depth interpretation of the model based on the quality of the features and the variation of the imposed rank.
Document type :
Preprints, Working Papers, ...
Complete list of metadata
Contributor : Mohamad Jouni Connect in order to contact the contributor
Submitted on : Thursday, December 30, 2021 - 2:45:35 PM
Last modification on : Monday, April 4, 2022 - 9:28:32 AM


Files produced by the author(s)


  • HAL Id : hal-03480890, version 2


Mohamad Jouni, Mauro Dalla Mura, Lucas Drumetz, Pierre Comon. Multifeature Hyperspectral Unmixing Based on Tensor Decomposition. 2021. ⟨hal-03480890v2⟩



Record views


Files downloads