Effect of Spectral Pre-Processing Methods on Non-Destructive Quality Assessment of Oranges Using NIRS

Near-infrared spectroscopy (NIRS) combined with chemometric methods of spectral pre-processing and multivariate modeling is popular non-destructive method for measuring quality attributes of fruits and vegetables. The present study investigated the feasibility of reflectance NIRS in a spectral range of 930 1650 nm for non-destructive prediction of soluble solids content (SSC) and titratable acidity (TA) in Thomson oranges. The effect of spectral pre-processing methods on the accuracy of multivariate predictor models was also assessed. Partial least squares (PLS) multivariate calibration models were developed using the reference measurements and pre-processed spectra. The following methods were used: smoothing (moving average (MA), Savitzky-Golay (SG), wavelet transform (WT)); normalizing (multiplicative scatter correction (MSC), standard normal variate (SNV)); and increasing the spectral resolution (the first and second derivatives (D1, D2)) to predict the SSC and TA of oranges. The results indicate that reflectance NIRS plus chemometrics gives the potential for non-destructive prediction of SSC and TA in oranges. Preprocessing methods directly affected the results of the PLS models. The best prediction results for SSC (RMSEC = 0.430, rc = 0.923, RMSEP = 0.451, rp = 0.936, SDR = 2.798) and TA (RMSEC = 0.133, rc = 0.883, RMSEP = 0.177, rp = 0.863, SDR = 1.853) were achieved based on a combination of the MA + SNV pre-processing methods.