Tracking of chemical gas plumes in hyperspectral video sequences

Hyperspectral imagery is nowaday widely used in numerous image processing fields. This imagery technique simultaneously acquires up to several hundreds of images of a same scene at different spectral wavelengths and stack them all in a data cube. Each pixel is therefore no longer a triplet of values as it is the case in classical RGB imagery, but a n−dimensional vector corresponding to a reflectance spectrum (see figure 1). This allows a fine description of the spectral properties of the scene, leading to a better identification of physical materials composing it. Hyperspectral imagery has a number of real-life applications in various remote sensing fields such as vegetation mapping, geological and hydrological sciences as well as food quality inspection and medical imagery. However, hyperspectral imagery remains an active research field due to all the issues rising when applying classical image processing operations (such as segmentation, denoising, mathematical morphology) and proper hyperspectral operations (spectral classification, spectral unmixing, etc) on hyperspectral images. It is now possible to acquire hyperspectral video sequence at near real-time rate thanks to the fast development of imaging sensors. The combination of a fine spectral description