Perspective of inline control of latent defects and diseases on french fries with multispectral imaging

In this paper, the feasibility is investigated to improve discrimination between different defect and diseases on raw French fries with multispectral imaging. Four different potato cultivars are selected from which French Fries are cut. Both multispectral images and RGB color images are classified with linear Bayes normal classifier and a support vector classifier. The effect of applying different preprocessing techniques on the spectra prior to classification was also investigated. The classification result are compared with both RGB images and the full spectra classification results. Experimental results indicate that the support vector classifier gives the best performance for both multispectral and RGB color images and is less preprocessing dependent. The multispectral image classification results outperform the RGB color classification results with a factor 15 at best. An explorative multispectral analysis also shows that latent defects can be detected with multispectral imaging, in contrast with traditional color imaging.