Comparison of Data Pre-processing in Pattern Recognition of Milk Powder Vis/NIR Spectra

The effect of data pre-processing, including standard normal variate transformation (SNV), Savitzky-Golay first derivative transformation (S. Golay 1st-Der) and wavelet transforms (WT) on the identification of infant milk powder varieties were investigated. The potential of visible and near infrared spectroscopy (Vis/NIRS) for its ability to nondestructively differentiate infant formula milk powder varieties was evaluated. A total of 270 milk powder samples (30 for each variety) were selected for Vis/NIRS on 325-1075 nm using a field spectroradiometer. Partial least squares (PLS) analysis was performed on the processed spectral data. In terms of the total classification results, the model with the wavelet transforms processed data is the best, and its prediction statistical parameters were r2 of 0.978, SEP of 0.435 and RMSEP of 0.413. This research shows that visible and near infrared reflectance spectroscopy has the potential to be used for discrimination of milk powder varieties, and a suitable pre-processing method should be selected for spectrum data analysis.