Nondestructive detection of zebra chip disease in potatoes using near-infrared spectroscopy

Abstract

Near-Infrared (NIR) spectroscopy (900–2600 nm) was evaluated as a rapid, non-destructive method for detection of zebra chip disease (ZC) in potatoes. Two models were tested; one that directly correlated spectra with ZC and one that measured sugar concentrations which in turn are known to be correlated with ZC. Applying stepwise regression in conjunction with canonical discriminant analysis to raw spectra, total classification accuracy of 98.35% was achieved in discriminating infected potatoes from control, with 2% false negative and 1% false positive error rates. The same analysis applied to 2nd derivative spectra yielded 97.25% accuracy with equal false negative and false positive error rates. Canonical discriminant analysis applied to sucrose, glucose, and fructose concentrations previously determined by high-performance liquid chromatography yielded 96.7% classification accuracy, with 4.3% false positive and 2.3% false negative rates. Accuracy did not significantly differ when fructose was excluded from the model. Partial least squares regression models built to predict sugar concentrations from the 2nd derivative NIR spectra resulted in R2 for actual vs. predicted concentrations of 0.7 and 0.72 respectively for sucrose and glucose, 0.63 for fructose, and 0.81 for total sugars. Given the relatively low R2 values in measuring sugar concentrations directly from the spectra it was concluded that classification accuracy is highest for models that directly correlate spectral features to ZC without considering sugar concentrations. Furthermore, this indicates that although NIR can detect infection, it may not be effective for evaluating severity of ZC in fresh potatoes.