Study On Identification Model Of Geographic Indication Rice Adulteration Based On SVM And KNN Algorithm

Jilin Province’s geographical indication rice is famous all over the country.Illegal traders make use of the commercial and nutritional value of landmark rice to make huge profits,which makes adulteration of rice emerge in endlessly.In recent years,the safety of rice has attracted much attention from the society,especially in the adulteration of geographical indication rice.This paper mainly discusses whether the geographical indication rice is mixed with ordinary rice,and the feasibility of using machine learning method to identify,which can provide an effective basis for the quality protection system of geographical indication rice.In this study,the main research object is the adulterated rice from Qianguo County,Songyuan City,Jilin Province and the common rice mixed with different proportions(2%,4%,6%,8%,10%,12%,14%,16%,18%,20%,25%,30%)in Jinting Town,Suzhou City,Jiangsu Province.The data fusion method of near infrared spectroscopy and mineral element fingerprint technology is adopted.Combining SVM model,KNN model and Adaboost-SVM model improved by Adaboost algorithm,Adaboost-KNN model are four machine learning methods to establish a fast identification model for rice adulteration.Principal component analysis is used to extract the principal component characteristics of near infrared spectral data under Savitzky-Golay convolution smoothing pretreatment to obtain spectral data of three principal components;Through the importance ranking mechanism based on Gini index in random forest algorithm,13 mineral elements are ranked,and 13 multi-element subsets of mineral element data are generated.Finally,using the intermediate fusion data of the two data as the identification data of landmark rice and adulterated rice,the model is established and optimized,and the model established by the research institute is evaluated and compared by 4-fold cross-validation and confusion matrix.The main conclusions are as follows:(1)Adulteration identification model based on SVM model and Adaboost-SVM model.Before optimization,the overall accuracy of Adaboost-SVM model is higher than that of SVM model.The generalization ability and accuracy of the two models after optimization are higher than that before optimization.Both models before and after optimization can effectively identify landmark rice and adulterated rice mixed with different proportions.When the adulteration ratio is 6% and 30%,the accuracy of both models is 100%.(2)Adulteration identification model based on KNN model and Adaboost-KNN model.Before optimization,the overall accuracy rate of Adaboost-KNN model is higher than that of KNN model.The generalization ability and accuracy rate of the two models after optimization are higher than that before optimization.Both models before and after optimization can effectively identify landmark rice and adulterated rice mixed with different proportions.When the identification ratio is 6% and 30%,the identification effect is the best,and the accuracy rate of both models is 100%.(3)SVM model,Adaboost-SVM model,KNN model and Adaboost-KNN model have the same accuracy of 100% when the discrimination ratio is 6% and 30%.The minimum detection limits of the four models are 2%,and the accuracy rates are 100%,97.75%,97% and 100%,respectively.(4)In the research of distinguishing landmark rice from adulterated rice with adulterated ratio of 2%,the models are compared.From the aspect of accuracy and generalization ability of the models,Adaboost-KNN model and SVM model,Adaboost-SVM model and KNN model are ranked from high to low.Considering the cost of model construction,Adaboost-SVM model,Adaboost-KNN model,SVM model and KNN model are arranged from high to low.(5)The application of data fusion technology combined with SVM model,KNN model,Adaboost-SVM model,Adaboost-KNN model four machine learning methods is feasible in the identification of adulteration of landmark foods.