A spectral graph based approach to analyze hyperspectral data

Hyperspectral imaging has emerged as a promising tool in the identification of objects and the state of objects, by their chemical and material composition. Hyperspectral imaging acquires spectral information at each pixel location across a wide range of the light spectrum. This enhanced spectrum information also comes with additional noise including spectral mixing, blurring and acquisition distortions. The analysis and processing of this high dimensional data requires efficient specialized techniques. We discuss a new novel graph based method for dimension reduction, image segmentation and classification based on the Ginzburg-Landau functional from classical PDE minimization. It aims to efficiently preserve as much spectral and structural information as possible. We show results from a field test of a wide field of view imaging spectrometer (WFIS) high performance hyperspectral imager designed for atmospheric chemistry and aerosols measurement from aircrafts and satellites.