Abstract: In this study, we proposed a high-speed Raman imaging method for the identification of adulterants in milk powder. In this method, a novel self-encode shrinkage neural network (SSNN) was developed to extract intrinsic information from the low signal-to-noise ratio Raman image with short integration time. Thereafter, multivariate regression models for quantitating the adulterant content in milk powder accurately were developed with the SSNN filtered Raman images. The coefficient of determination (R2) of these quantitative models for various adulterated samples was above 0.95. Through this method, a sample region size of 50 mm × 50 mm could be scanned with Raman imaging technique within two minutes, 30 times faster than traditional Raman imaging method. These satisfactory results demonstrate that this method can successfully meet the demand of milk powder adulteration detection in practice and can be used to detect adulteration in other non-homogeneous food systems.

Key words: milk powder adulteration; high-speed Raman imaging; self-encode shrinkage neural network; extraction of intrinsic signals