Research On Apple Identification Method Based On Hyperspectral Technology And Chemometrics Method

Apple is one of the most nutritious fruits in the world.In recent years,with the development of regional brands,Apple counterfeits have emerged one after another.In this study,Apple samples from Aksu,Gansu,Henan and Shandong were collected for hyperspectral images and chemical data,the study of Apple Identification Method based on hyperspectral technology and Chemometrics.In this study,Apple’s identity is divided into two attributes:the identification of place of production and the identification of apple varieties.The information of variety and producing area is fused to identify apple accurately.The specific content and results of the study are as follows:(1).In this study,the same apple cultivar identification model and different apple cultivar identification model are studied,and the KNN,SVM and PLS prediction models are established after different spectral data processing.The results show that the KNN model is the best for the same origin,the highest accuracy of prediction set is 100%,The PLS＿DA prediction result is poor,and the SVM modeling result is the worst,and the PLS model is the best for the different origin,the highest accuracy of prediction set is 99%.(2).According to the variety information,the same variety and different variety of apple were studied,and the KNN,SVM and PLS＿DA prediction models were established after different spectral data processing.The results showed that the PLS model had the best prediction results for different varieties and the same varieties,and the highest prediction accuracy is 100%.(3).In this study,PLSR was used to model the preprocessed data.The results show that SD,MSC and SNV are the best preprocess methods,the correlation coefficient R of each pretreatment method is more than 0.82.The result shows that MSC-SD combined with KNN is the best way to predict the producing area,the prediction accuracy is 98.68%,SD combined with PLS＿DA was the best method for variety prediction,and the prediction accuracy is 99.6147%.(4).In this study,SPA,CARS and PCA loading method are used to screen the characteristic bands of the original spectral data.The results show that the characteristic band modeling is less accurate than the whole band prediction of the original spectrum.The prediction results are more accurate when the reasonable values are used to screen the characteristic bands,in which CARS is the best,and the accuracy of the modeling prediction set is more than 80%(5).Apple identification model was established by fusing origin identification and variety identification.The prediction accuracy of prediction set is 97.42% and that of correction set is98.77%.The results show that the model of fusion variety identification and origin identification is effective for apple identification,and can identify apple accurately.