Classification of Hyperspectral Images as Tensors Using Nonnegative CP Decomposition

Mohamad Jouni (1) , Mauro Dalla Mura (1, 2) , Pierre Comon (3)

1 GIPSA-SIGMAPHY - GIPSA - Signal Images Physique
2 Tokyo Tech World Research Hub Initiative (WRHI), School of Computing, Tokyo Institute of Technology, Tokyo
3 GIPSA-CICS - GIPSA - Communication Information and Complex Systems

A Hyperspectral Image (HSI) is an image that is acquired by means of spatial and spectral acquisitions, over an almost continuous spectrum. Pixelwise classification is an important application in HSI due to the natural spectral diversity that the latter brings. There are many works where spatial information (e.g., contextual relations in a spatial neighborhood) is exploited performing a so-called spectral-spatial classification. In this paper, the problem of spectral-spatial classification is addressed in a different manner. First a transformation based on morphological operators is used with an example on additive morphological decomposition (AMD), resulting in a 4-way block of data. The resulting model is identified using tensor decomposition. We take advantage of the compact form of the tensor decomposition to represent the data in order to finally perform a pixelwise classification. Experimental results show that the proposed method provides better performance in comparison to other state-of-the-art methods.

Domaines: Traitement du signal et de l'image [eess.SP]

Mohamad Jouni, Mauro Dalla Mura, Pierre Comon. Classification of Hyperspectral Images as Tensors Using Nonnegative CP Decomposition. ISMM 2019 - 14th International Symposium on Mathematical Morphology, Jul 2019, Saarbrücken, Germany. pp.189-201. ⟨hal-01998121⟩