ENBIS-18 in Nancy

2 – 25 September 2018; Ecoles des Mines, Nancy (France) Abstract submission: 20 December 2017 – 4 June 2018

Simultaneous Sparse Approximation for Variable Selection and Classification. Application to Near Infrared Spectra and Hyperspectral Images of Wood Wastes

3 September 2018, 14:20 – 14:40


Submitted by
David Brie
El-Hadi Djermoune (Université de Lorraine, CNRS), Thomas Aiguier (Université de Lorraine, CNRS), Cédric Carteret (Université de Lorraine, CNRS), David Brie (Université de Lorraine, CNRS)
We propose a simultaneous variable selection method for material sorting based on near infrared spectroscopy. The objective is to perform quick classification in industrial wood recycling processes based only on a few spectral bands. The spectra are first jointly modeled as linear combinations of explanatory variables drawn from a collection of wavelengths (dictionary). The aim is to select a common subset of variables shared by several spectra. The selection is then addressed using a simultaneous sparse approximation method in which the coefficients related to different spectra are encouraged to be piecewise constant, i.e. the coefficients associated to successive spectra should have comparable magnitudes. We propose also a nonnegative version of the optimization problem in which the magnitudes are also constrained to be positive. These problems are solved using the fast iterative shrinkage-thresholding algorithm. The proposed approaches are illustrated on a dataset of 294 infrared spectrometry measurements of wood wastes containing 1647 wavelengths. We show that the selected variables lead to better classification performances as compared to standard approaches. Finally, we show how these methods naturally extend to hyperspectral image classification for pushbroom imagers. Results obtained on real hyperspectral data of wood wastes confirms the effectiveness of the method.

Return to programme