Abstract:
Abstract: As a scattering spectrum, Raman spectroscopy has some advantages in non-invasive detecting. The hyperspectral data contain not only conventional image but also spectral information in each pixel. In this study, a line-scanning Raman hyperspectral imaging system was built to detect and quantify the melamine mixed in the milk powder with large sample areas in a fast and nondestructive way. The Gaussian filter smoothing and an adaptive iteratively reweighted penalized least squares (air PLS) method were used to remove noise signal and fluorescence interference. The corrected images at 671.71 cm-1 waveband were extracted for detecting the melamine in the milk powder. Firstly, the penetration depth of Raman signal produced by melamine in the milk powder was measured. A designed two-layer sample was applied to measure the Raman signals after passing through milk layers of different thicknesses. According to the results, the optimum thickness of mixed samples was set to be 2 mm. Then, melamine-milk mixtures with 10 different concentrations were prepared for the experiment. Each mixture was collected by a designed aluminium alloy container with a sample thickness of 2 mm. In this case, the melamine particles at the bottom of mixed sample could be collected. After data preprocessing, a linear analysis of the averaged Raman intensity of each pixel was performed, and the concentration and distribution information of the melamine particles were finally obtained using a simple binarization arithmetic in the single-band image of mixtures at 671.71 cm-1 waveband. The results showed that there was a linear relationship between the melamine concentration and the average Raman intensity of all pixels in the region of interest of the corrected image at 671.71 cm-1 waveband, and the coefficient of determination was 0.9954. In the binary images, the number and spatial location information of melamine particles could be visually displayed. Meanwhile, the total number of the additive pixels increased nonlinearly. It meant that the binary images from this research represented the accumulation of multiple layers in sample. At low concentrations, the Raman signal generated from the additive particles at the sub-surface is too weak to detect. When the additive concentration increases to a certain degree, the Raman signal generated from the additive particles at the sub-surface can be collected. In these areas, the pixels are identified as additive pixels even if there is no additive particle at corresponding surface. This situation led to a significant increase in the number of additive pixels. The research demonstrates that the Raman intensity in single-band corrected images can be used for quantitative analysis of melamine, and the binary images can reveal the identification and the distribution of melamine particles in the skim milk powder. More Raman active additives in powdered food could be detected in the same way. In our research, the milk powder samples can be scanned directly without any chemical reagents. The process of converting to liquid is dispensable. The limit of detection for melamine concentration was estimated as 0.01% with a total detection area of 40 mm × 80 mm each time. The results show that the line-scanning Raman hyperspectral imaging system has shown a great potential for rapid and non-invasive measurement of samples with large areas.