Research On Rapid Detection Technology For Food Packaging Plastic Based On NIR

Food safety has become a social issue in nowadays as significantfood safety incidents frequently hit us. The blasting fuse of incidents is notonly from food quality, while packaging plastic has turned into a potentialthreat via processing, storing and salling links. The demand of plastic forfood packaging is over600million tons per year, while physical andchemical method is main detecting facility, which is not safe and efficientenough. So a highly accurate and efficient method is needed for port ofimport and export.In this paper, we concentrated on5typical food packing imported raw plastics:polyethylene （PE）, polypropylene （PP）, polystyrene （PS）, polyethylene terephthalate（PET） and acrylonitrile butadiene btyrene （ABS）. Combined Chemometrics and NIR,qualitative and quantitative models have been set.The main research of this paper contents includes:Firstly, in-depth research on the current situation of domestic and foreign aboutplastic detecting method and the NIR used in this field has been done. Comparison ofmodelling algorithm has been listed in the paper, and the characteristics andinfluence factors of our model analysis have been calculated.Secondly, before modelling, spectrums must be of rich information and lessnoise. Realted factors are diffuse reflection model, scanning resolution, scanningfrequency, sample rotating model and loading weight. So comparison experimentshave been made to come up the best choice.After that, in the qualitative part, factor analysis was used to realize datacompression and feature extraction. A new factor chosen assessment was proved inthis paper which was based on mahalanobis distance, and confirmatory experimentshave been made. In quantitative part, the main algorithm is LS-SVM, with whichverification was done.Lastly, the mixed experiments were made to verify the model effect and results showed that classification of the former4types of plastics and trademarks weresuccessful, while the density classification failed to differ from each other. And inquantitative part, LS-SVM algorithm was used to predict nitrogen potency of ABS.With second derivative and wave range from5000cm^-1to5400cm^-1, the modelfunctiones well, whose R²is94.49, RPD is4.26, SEVC of training samples is0.45and the SEP of testing samples is0.26, which achieved better result than OPUS.