External quality grading of Jonagold apples using a multi-spectral vision system

This paper presents a method to sort Jonagold apples using a four bands multi-spectral image acquisition device. Multi-spectral images of sound tissue and various defects were acquired. Defects could be divided into four classes: slight defects (e.g. small russet), more serious defects (scar tissue), defects leading to the rejection of the fruit (e.g. rot) and recent bruises (between one hour and two hours old). Image segmentation was based on the Bayes' theorem. Each pixel of the fruit was classified into 'healthy' or 'defect' classes according to the probability distribution of the spectral components of each class. Once segmented, the fruit was graded by linear discriminant analysis on the basis of the relative area of the defect and statistical parameters computed on the spectral components of the two tissues classes. Results (cross validation) showed 94% and 84% if the sound and defective fruits respectively well classified. Most of the misclassified defective fruits (89%) belonged to the slight defect category.