Spatial extensions to self - modeling curve resolution

Through recent developments in data acquisition procedures, large amounts of spatially resolved hyperspectral data have become available. However, the development of adequate data analysis methods which exploit the intrinsic spatial information lags behind. We propose a self-modeling curve resolution (SMCR) algorithm which takes spatial relationships into account. This is accomplished by enforcing spatial smoothness constraints during alternating least squares (ALS) estimation. The standard columnwise approach to this problem bears significant difficulties with respect to stability and and convergence of the ALS algorithm. Therefore, a superior resolution method based on multiple linear regression has been developed. We show that smoothness constrained estimates may increase the physical or chemical interpretability significantly, although the possible gain is dependent on the structure of the data. 2 Chapter 1