Abstract:
This research aims to apply near infrared spectroscopy (NIRS) for the detection of total fungal, ochratoxigenic Aspergillus and ochratoxin A contamination in green coffee beans. Quantitative models for predicting moisture content, fungal contamination and ochratoxin A contamination in green coffee bean samples were developed from the correlation between laboratory data (percentage of moisture content, percentage of total fungal infection, percentage of Aspergillus section Nigri infection and percentage of Aspergillus section Circumdati infection) and optical data (raw spectra and mathematically pretreated spectra) from NIR scanning on the green coffee bean samples by using the method of partial least square regression (PLSR). The best model for predicting moisture content was developed from the mean normalization pretreated spectra, with the coefficient of correlation (r) of 0.970, standard error of prediction (SEP) of 0.176% and bias of -0.012%. The best model for predicting total fungal contamination was developed from the range normalization pretreated spectra (r = 0.835, SEP = 15.205%, bias = 0.718%). The best model for predicting Aspergillus section Nigri contamination was developed from the multiplicative scatter correction pretreated spectra (r = 0.865, SEP = 19.051%, bias = -1.478%). The best model for predicting Aspergillus section Circumdati contamination was developed from the second derivative by Savitzky-Golay method of 21 points pretreated spectra (r = 0.972, SEP = 7.704%, bias = 0.351%). For qualitative models, the classification models of Aspergillus section Circumdati contamination in green coffee bean samples developed using partial least square-discriminant analysis (PLS-DA) provided the highest percentage of overall correct classification of 100%. The quantitative models analysis of ochratoxin A contamination were developed from the correlation between chemical analysis data (ochratoxin A concentration) and optical data (raw spectra and mathematically pretreated spectra) obtained using NIR scanning on 3 types of green coffee bean samples (whole green coffee beans, grounded green coffee beans and crude extract from green coffee beans) by using the method of PLSR. The best model for predicting ochratoxin A contamination in green coffee bean samples was developed from optical data of whole green coffee bean samples with the first derivative by Savitzky-Golay method of 21 points pretreated spectra, with the r, SEP and bias of 0.814, 1.965 µg/kg and 0.358 µg/kg, respectively. For qualitative analysis, the classification models of ochratoxin A contamination (≤5 and >5 µg/kg) developed using PLS-DA provided the highest percentage of overall correct classification of 100%.