Milk powder authenticity is one of the critical issues in food safety,which non-directional screening remains a hotspot and frontier in analytical science.A robust model driven Raman hyperspectral imaging method(RMD-RHIM) for the non-directional screening of milk powder authenticity was proposed by utilizing the spectral features and spatial features of milk powder samples efficiently,which enabled RMD-RHIM to identify the unknown adulterants in milk powder as outliers,thus avoiding to estimate the uncertainty of numerous adulterants.In RMD-RHIM,the adaptive iterative reweighted penalized least square(airPLS) was adopted to suppress the background information in Raman spectra,and then modified iterative reweighted partial least square(mIRPLS) was used to detect outliers in thousands of Raman spectra.As a result,RMD-RHIM was capable of identifying any adulterant in milk powder as an outlier,which was illustrated as a distorted pix in binary image.The classification accuracies of RMD-RHIM for positive and negative samples were 98.3% and 93.3%,respectively.The results obtained indicated that RMD-RHIM was a promising tool for the non directional screening of milk powder,which may be well extended to other food systems.