Sorting of In-Shell Pistachio Nuts from Kernels Using Color Imaging

Sorting pistachio kernels from in-shell nuts currently requires a combination of automated and manual sorting, an expensive and labor-intensive two-stage process. This research demonstrates the feasibility of using color imaging as a basis for distinguishing both regular and small in-shell pistachio nuts from kernels in the pistachio nut process stream. Two algorithms were developed to classify images of in-shell nuts, small in-shell nuts, and kernels. The first algorithm used a discriminant analysis (DA) routine to evaluate features extracted from the images based on histograms of red, green, and blue (RGB) pixel intensities, and resulted in a 99.9% overall accuracy for separating regular in-shell pistachio nuts from kernels. Small in-shell pistachio nuts were harder to discriminate from kernels, with an overall accuracy of 85%. The second algorithm used a k-nearest neighbors (knn) routine to evaluate features based on color histograms plus intensity slope information. The knn routine matched the accuracy of the DA algorithm for distinguishing regular in-shells from kernels with 99.9% correct. For the small in-shell case, however, the knn approach was superior with 96% accuracy. When used in a high-speed color imaging system, the algorithms would provide the means for economical high-speed sorting of in-shell pistachio nuts and kernels.