A new hyperspectral unmixing method using co-registered hyperspectral and panchromatic images

Simon Rebeyrol (1, 2) , Yannick Deville (1) , Véronique Achard (2) , Xavier Briottet (2) , Stéphane May (3)

1 IRAP - Institut de recherche en astrophysique et planétologie
2 DOTA, ONERA, Université de Toulouse [Toulouse]
3 CNES - Centre National d'Études Spatiales [Toulouse]

In this paper, we propose a novel way of unmixing space-borne hyperspectral images using a co-registered hyperspec-tral image and a high-spatial-resolution panchromatic image. The unmixing is achieved by a two-step approach. First, an extraction of pure pixels using a panchromatic homogeneity criterion is performed and a reconstruction error map is generated. Then, a local semi-supervised Non-negative Matrix Factorization (NMF) is performed on badly reconstructed areas in order to estimate material spectra that are not represented by pure pixels in the image. This approach is tested on real data for the first step, and on both synthetic and real data for the second step.

Mots clés: HYPERSPECTRAL; PANCHROMATIC; UNMIXING; SPACEBORN

Domaines: Sciences de l'ingénieur [physics]; Physique [physics]

Simon Rebeyrol, Yannick Deville, Véronique Achard, Xavier Briottet, Stéphane May. A new hyperspectral unmixing method using co-registered hyperspectral and panchromatic images. Whispers 2019, Jan 2019, AMSTERDAM, Netherlands. ⟨hal-02409424⟩