DETECTION OF PITS IN TART CHERRIES BY HYPERSPECTRAL TRANSMISSION IMAGING

The presence of pits or pit fragments in pitted cherry products poses potential hazard to consumers and thus make the food industry liable for economic losses. The objective of this research was to develop a hyperspectral imaging technique for detecting pits in tart cherries. Hyperspectral transmission images were acquired from 'Montmorency' tart cherries for four orientations over the spectral region between 450 nm and 1,000 nm before and after pits were removed. Cherries of three size groups (small, medium, and large), each with two color classes (light red and dark red) for two harvest dates, were used for determining the effect of fruit orientation, size, and color on pit detection. Additional cherries were bruised and then subjected to two different post-bruising treatments (room storage vs. cold storage) to study the bruising effect on pit detection. A feed-forward backpropagation neural network (NN) classifier was developed to classify cherries with and without pits. Single spectra and selected regions of interest (ROIs), covering the spectral region between 692 nm and 856 nm, were compared as inputs for the NN. ROIs resulted in 3.5% error in incorrect classification of cherries with pits and 3.1% error for cherries without pits, which were less than half of those from single spectra, when all cherries of mixed groups were used. Fruit size and defect had a great effect on pit detection; the false negative error (incorrect classification of cherries with pits) increased dramatically when small or defective (bruised) cherries were not included in the NN training. However, the effect of fruit orientation or color on NN classifications was either small or negligible.