Hyperspectral Image Unmixing: Accounting For Wavelength Dependence∗

We introduce a method for hyperspectral unmixing that incorporates wavelength dependence in addition to spatial dependence. Spatial dependence is incorporated into the model using class labels on the pixels that is assigned using spectral clustering. Wavelength dependence is introduced by correlating the errors in the unmixing regression models. We propose a non-standard alternating direction method of multipliers (ADMM) algorithm to solve the resulting non-convex optimization problem that simultaneously recovers the abundances and the sparse precision matrices of the spectral signatures. Using data collected by the SpecTIR imaging sensor, we show that the proposed method outperforms several other well-established unmixing models.