Classification of intact macadamia nuts using NIR spectroscopy

Introduction: The major macadamia (Macadamia integrifolia) nut defects can be related to insect damage, mould growth, germination and discoloration. These defects are assessed visually and this procedure is not accurate. As NIR spectroscopy has been implemented to assess quality parameter in various food produces, the objective of this study was to evaluate the feasibility of NIR diffuse reflectance spectroscopy to classify intact macadamia nuts based on their defects.
Materials and Methods: A total of 100 intact macadamia nuts were collected in 2014 at the end of the harvest season. The nuts (n=25) were sorted based on their quality attributes, as such: i. sound nuts; ii. nuts with insect damage caused by Ecdytolopha aurantiana; iii. nuts with insect damage caused by Leucopteara coffeella; iv. cracked nuts caused by seed germination. All nuts were harvest in a commercial orchard located in Jaboticabal, SP, Brazil. After temperature stabilization (~25°C), the spectra were collected by using a FT-IR spectrophotometer (Spectrum 100N, PerkinElmer) in the diffuse reflectance mode over the range of 4,000-10,000 cm-1 on the nut surface of two different positions (64 scans, spectral resolution of 2 cm-1). The classification of the macadamia nuts into one of the classes was accessed using PCA-LDA procedure. It was also studied the classification and prediction of the nuts stated as good (without defects) and bad (with any defects) using PCA-LDA and PLS-DA techniques. Kennard–Stone algorithm was applied to divide the data into calibration and validation set. Different pre-processing was applied to spectra, namely standard normal variate (SNV) and second derivative of Savitzky-Golay.
Results and Discussion: The best PCA-LDA classification result for the different defects was obtained when the spectra were pre-processed using SNV followed by the second derivative of Savitzky-Golay and elimination of 8 outlier which were detected by the Hoteling T2 (P<0.050) test. It was possible to get 93.2% accurate classification with this procedure. The observed sensibility values were 98, 100, 87, and 94% for “good”, “bad”, “borer” and “crack” macadamias, respectively. To classify the macadamia nuts in good or bad the PCA-LDA and PLS-DA models have the accuracy of 88.3% and 87.7% for calibration set, respectively. The accuracy reduced to 80% and 86.7% when the prediction was carried out using an external validation set. The sensibility and specificity values for the prediction set were 85, 90% and 73, 83%, respectively for “good” and “bad” nuts.
Conclusion: The NIR spectroscopy can be successfully used to classify macadamia nuts based on their defects. This technique can be used as a non-destructive method and could be valid and simple tool to reduce the quality control costs of monitoring macadamia nuts’ quality.
Summary: PCA-LDA models on NIR spectral data (4,000-10,000 cm-1) illustrated that this technique can be used as a non-destructive method to classify intact macadamia nuts based on their defects, and have a considerable potential for monitoring nuts’ quality in grading line.