Abstract:Thisstudy aimsto explore a method for the fast and nondestructive detection of[Nε-（1-carboxymethyl）-L-lysine（CML）].Hyperspectral imaging technology was employed to detect 8 kinds of homemade fried potato chips and the average spectral value of each hyperspectral image was extracted as the feature parameter.Meanwhile，the content of CML was determined by liquid chromatography-mass spectrometry（LC-MS）.The suitable spectral pretreatment and modeling method were then explored.Firstly，the black-and-white correction was carried out for hyperspectral images.Secondly，standard normal variable transformation was used to eliminate the influence of solid particles，scattering and optical path change on spectrum.Thirdly，the average spectral reflectance values from images at the 200th-1 000thwavebands were taken as the characteristic parameters for the building of three prediction models based on principal component regression，partial least squares regression and BP neural network，respectively.The comparison demonstrated that BP neural network had the prediction accuracy of 99.67%，the determination coefficient of 0.99 and the root mean square error of 0.22.Furthermore，five groups of training set and prediction set were randomly selected to verify the model robustness.The result showed that the average prediction accuracy，determination coefficient and root mean square error was 96.23%，0.99 and 0.22，respectively.Therefore，it is feasible to combine the hyperspectral imaging technology with BP neural network for the rapid prediction of CML content in fried potato chips.