Abstract: In this paper, a moving window spectral angle mapping-based Raman hyperspectral imaging method (MWSAM-RHIM) was proposed for non-targeted screening for milk powder safety, which could effectively compensate for the shortcomings of traditional targeted detection methods. In MWSAM-RHIM, an improved spectral angle mapping algorithm was used to calculate the spectral similarity of unknown samples. The pixels beyond the threshold were marked as adulterated to generate a visual binary image. The results showed that the recognition rates of negative and positive samples by MWSAM-RHIM method both reached 93.3%, indicating that this method can meet the needs of non-targeted screening for safety of dairy products and provide a new idea for rapid screening of other food samples.

Key words: Raman hyperspectral imaging, milk powder safety, spectral angle mapping, non-targeted screening, food safety