Detection and classification of latent defects and diseases on raw French fries with multispectral imaging

Abstract

This paper describes an application of both multispectral imaging and red/green/blue (RGB) colour imaging for the discrimination between different defect and diseases on raw French fries. Four different potato cultivars generally used for French fries production are selected from which fries are cut. Both multispectral images and RGB colour images are classified with parametric and non-parametric classifiers. The effect of applying different preprocessing techniques on the spectra was also investigated. The best classification results in terms of accuracy, yield and purity are obtained with a modified version of standard normal variate (snv_mod) preprocessing for different classifiers and potato cultivars. The classification results of the multispectral images are compared with RGB images. The results show that the support vector classifier gives the best classification performance for the snv_mod preprocessed multispectral images and k-nearest neighbours classifier gives the best classification performance for raw RGB images. The detection of the latent greening defect in French fries with the exploration of multispectral images shows the additional value of multispectral imaging for French fries. A comparison between the multispectral images and the RGB colour images confirms this since this type of defect is not visible in the colour images. © 2005 Society of Chemical Industry.