Find out more about subscribing to add all events.
This talk addresses online processing of hyperspectral images (HSI) as provided by wiskbroom and pushbroom scanning systems widely used in industrial applications of HSI. First, the online deconvolution is considered and a least-mean-squares (LMS)-based framework accounting for the convolution kernel non-causality and including non-quadratic (zero attracting and piece-wise constant) regularization is proposed. This results in the so-called sliding block regularized LMS (SBR-LMS) which maintains a linear complexity compatible with real-time processing in industrial applications. Then, online blind unmixing of HSI is addressed as a Non Negative Matrix Factorisation (NMF) problem. To face the non-uniqueness of the NMF, 2 different regularisation terms are considered. The first one is a minimum volume term and the corresponding criterion is minimized using alternate minimisation with multiplicative updates. The second one is a minimum distance term and the criterion minimisation is achieved by an ADMM approach.
The pro and cons of both approaches are discussed and an application to wood singularity detection is presented.
David Brie completed his Ph.D. degree in 1992 and was awarded the "Habilitation à Diriger des Recherches” certificate in 2000, both from the University of Lorraine, Nancy, France. He is currently full professor at the Department of Telecommunications and Networking of the Institut Universitaire de Technologie, University of Lorraine, France and is member of CRAN laboratory (UMR 7039 CNRS). His research interests mainly concern inverse problems and multidimensional signal processing. Since 2013, he is member of the board of the GRETSI association and is the editor-in-chief of the French journal Traitement du Signal. He will be the general chair (with J.Y. Tourneret) of the IEEE CAMSAP 2019.