Nondestructive Testing Of Meihe Rice Quality Based On Hyperspectral Image Technology

Geographical indications are main tags to evaluate the quality of agricultural products.Due to the advantages of growth environment and place of production,unique intrinsic quality of landmark rice has formed.Because Meihe rice has good geographical location(Gold rice belt)and high-quality water source(Huifa River),It has formed a high-quality index with its own characteristics.Meihe rice with high economic value has limited output,so there is full of adulterated rice and fake rice in the market.The situation not only damages the market order but also hurts the interests of consumers.Now information logistics has rapidly developed,how to ensure the quality of Meihe rice is not affected by other factors,which requires to establish an efficient and fast detection technology for Meihe rice.With the rapid development of modern science and technology and the rapid rise of the food industry,simple and efficient detection methods are used to replace complex and time-consuming traditional methods,which has become a hot research topic at present moment.Schematic of hyperspectral imaging is considered as one of the current non-destructive detection technologies.It has a more comprehensive observation on the internal quality of rice.In this paper,three varieties of rice(Daohuaxiang,Akitakomachi and Jijing 60)from four producing areas of Meihe River(Shuguang Town,Wanlong Township,Heishantou Town and Jile Township)were used as experimental samples.Schematic of hyperspectral imaging PLSR used to rapidly and non-destructively detect the physical and chemical indexes of Meihe rice to obtain the partial least squares regression prediction model of fatty acid content and protein content,which realized the rapid and non-destructive identification of Meihe rice quality.Later,this method was used to detect other components of rice inside to ensure the quality of Meihe rice.By analyzing the spectral data and physical and chemical indexes of meihe rice,this paper obtained more accurate information about the origin of varieties of meihe rice,so as to provide a basis for the later identification of the quality of meihe rice.At last,the identification of the origin of meihe rice was completed by using spectral data,so as to reduce the circulation of adulterated rice in the market.In order to realize content determination of hyperspectral on fatty acid and protein in rice,rice samples after ridging and roughening were randomly placed on the blackboard of the stage to collect hyperspectral images according to a 3×5 grid.ENVI5.0 was used to extract the average spectral information in the region of interest.MATLAB2016 a was used to build a data model of rice.The spectra were used to combine the content of fatty acid and protein,which can obtain the prediction model of fatty acid and protein in Meihe rice.The optimal model of both was the partial least squares regression model.SPA characteristic bands were extracted from the spectral data.And the fatty acid and protein content prediction models were established by the obtained twenty-six characteristic bands and twenty-seven characteristic bands as independent variables.In order to realize more intuitive expression,In this paper,the spectral data of all pixels in the extracted hyperspectral image of rice are brought into the established prediction model of fatty acid and protein content,which can obtain the fatty acid and protein content of each pixel.The hyperspectral gray image is put into MATLAB for conducting pseudo-color processing.The distribution of fatty acid and protein content in rice is visualized.By using hyperspectral data to discriminate the origin and varieties of meihe rice,the accuracy of Fisher's discriminant model and BP neural network model in four production areas of different varieties is 86.30% and 93.40% respectively.The accuracy of Fisher's discriminant model and BP neural network model in four production areas of the same variety of rice is 95.80% and 97.10% respectively.The accuracy of Fisher's discriminant model and BP neural network model of meihe and non-meihe rice is 93.8% and 94.83% respectively.The accuracy of Fisher's discriminant model and BP neural network model of meihe rice is 97.30% and 97.70% respectively.The rapid and nondestructive discriminant analysis of rice varieties in meihe region has been realized.In order to clarify the distribution of nutrition components in meihe rice,the two-factor anova of varieties and origin of different varieties of rice from different production areas are carried out,and it is proved that the four physicochemical indexes(fatty acid,protein,ash and fat)are significantly different between varieties and origin.Distinguishing and analyzing the origin and varieties of rice from different origins and varieties were conducted,we may see that the distribution of physical and chemical indexes has formed its unique physical and chemical distribution characteristics because of the differences of the producing areas and varieties.A single physical and chemical index was used to distinguish rice varieties and producing areas,which has low accuracy.After that,four indexes were combined to distinguish and analyze.For the origin discrimination of Meihe rice from different producing areas of the same variety,the higher the discrimination rate indicated that the physicochemical index distribution of the variety of rice had its own characteristics in the producing area,which can more accurately judge the origin of Meihe rice.However,the discrimination of different varieties of rice from the same producing area,we can see whether the distribution of the physicochemical index of the variety in the producing area formed its unique characteristics.The higher the discrimination accuracy rate indicated that the variety factor in the producing area will have greater effect on the physicochemical index.