Automatic detection of object pixels for hyperspectral analysis

Abstract

A method is provided for automatically discerning between object and non-object pixels in a hyperspectral image data cube. In particular embodiments, the object of the method is a plant, plant part, plant trait, plant phenotype, plant pot or a plant medium. The method comprises a first step of providing a partial least squares discriminant analysis (PLSDA) algorithm and a second step of applying the PLSDA algorithm to a hyperspectral image data cube to automatically determine which pixels contain the spectral properties of the object. The PLSDA algorithm of the method can be generated by establishing a training matrix, performing an eigenvector decomposition of the training matrix, experimentally determining a weighted linear combination of object signal-containing eigenvectors, calculating a regression vector using the weighted linear combination of signal-containing eigenvectors, generating a mask matrix and multiplying the mask matrix by the hyperspectral image data cube along two spatial dimensions.