Detection of pits and pit fragments in fresh cherries using near infrared spectroscopy

Authors: Liang, Peishih; MOSCETTI, ROBERTO - University Of Tuscia; MASSINTINI, RICARDO - University Of Tuscia; Light, Douglas; Haff, Ronald - Ron

Submitted to: Near Infrared Spectroscopy Journal
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/15/2017
Publication Date: 7/10/2017
Citation: Liang, P., Moscetti, R., Massintini, R., Light, D.M., Haff, R.P. 2017. Detection of pits and pit fragments in fresh cherries using near infrared spectroscopy. Near Infrared Spectroscopy Journal. 25(3):196-202. https://doi.org/10.1177/0967033517712130.
DOI: https://doi.org/10.1177/0967033517712130

Interpretive Summary: NIR spectroscopy was evaluated as the basis for a rapid, non-destructive method for the detection of pits and pit fragments in fresh cherries. Statistical algorithms following various spectral pretreatments were applied to spectra of cherries with either no pit (NP), a whole pit (P), a half pit (HP), or a quarter pit (QP) to test various classification schemes. An iterative algorithm tested all combinations of pretreatments and parameters as input to the algorithm. In addition, a step forward feature selection algorithm was used to identify the most significant wavebands in order to isolate small sets (<10) of spectral bands that represent the entire spectra. The highest accuracy was achieved when the samples were combined into only two classes (NP vs. P+HP+QP) using all features (reflection at each wavelength) with no false positive error, 4% false negative error and 98% overall accuracy. Overall accuracy of the same model was reduced only slightly to 96% when employing only the four most significant features. Overall accuracy declined when models attempted to separate the classes of fragments, with the lowest being 92%, 83%, 86%, and 99% accuracy respectively in discriminating NP, QP, HP, and P classes separately. The high accuracy achieved using only four features indicates that reflection of light at specific NIR wavelengths is a suitable basis for high-speed, non-destructive detection of pits and pit fragments in cherries.

Technical Abstract: NIR spectroscopy in the wavelength region from 900nm to 2600nm was evaluated as the basis for a rapid, non-destructive method for the detection of pits and pit fragments in fresh cherries. Partial Least Squares discriminant analysis (PLS-DA) following various spectral pretreatments was applied to spectra of cherries with either no pit (NP), a whole pit (P), a half pit (HP), or a quarter pit (QP) to test various classification schemes. An iterative algorithm tested all combinations of pretreatments and parameters as input to the PLS-DA. In addition, a step forward feature selection algorithm was used to identify the most significant wavebands in order to isolate small sets (<10) of spectral bands that represent the entire spectra. The highest accuracy was achieved when the samples were combined into only two classes (NP vs. P+HP+QP) using all features (reflection at each wavelength) with no false positive error, 4% false negative error and 98% overall accuracy. Overall accuracy of the same model was reduced only slightly to 96% when employing only the four most significant features. Overall accuracy declined when models attempted to separate the classes of fragments, with the lowest being 92%, 83%, 86%, and 99% accuracy respectively in discriminating NP, QP, HP, and P classes separately. The high accuracy achieved using only four features indicates that reflection of light at specific NIR wavelengths is a suitable basis for high-speed, non-destructive detection of pits and pit fragments in cherries.