ABSTRACT
This chapter presents a clear and precise way the concept and importance of the segmentation stage within the hyperspectral image processing workflow. It focuses on elucidating the steps that can be applied to hyperspectral images that are related to food quality evaluation. The hyperspectral image segmentation methods are typically categorized as supervised and unsupervised. The difference from traditional image segmentation that is focused on a two- or three-dimensional space is that the methods for segmenting hyperspectral images must deal with many spectral channels. Pre-processing methods for hyperspectral images such as smoothing, multiplicative scatter correction, standard normal variate, Fourier transform, and Wavelet transform are utilized to calibrate the original spectra and improve the next steps, such as segmentation. Performance metrics provide support for interpreting and evaluating the accuracy of the method used in the segmentation. There are several metrics that can be used to quantify and provide support to a given method: Kappa coefficient, overall accuracy, average accuracy, and class-specific accuracies.
