Research On The Traceability Model Of Millet Origin Based On Hyperspectral

The “Food security issue” is related to social stability and the fundamental interests of public,and its significance is self-evident.Millet is one of the important food crops in my country.Due to the affecting factors such as climate and soil elements,the nutrient composition of millet in different production areas is different.Driven by economic interests,some illegal businesses forge the geographic source information of millet to disrupt market order,which increases the difficulty of monitoring the quality and safety of millet and other agricultural products.It also requires more towards the millet origin traceability methods and technologies at the same time.At present,the researches at home and abroad on the traceability of agricultural products such as wheat,beans,rice and other agricultural products has been very substantial.In terms of theoretical research and practical operation,it’s feasible to carry out the hyperspectral traceability on agricultural products.However,there is still little research on hyperspectral technology in the traceability of millet origin.It’s of great importance for China’s food security and the protection of agricultural geographical indications to develop an efficient and accurate traceability method suitable for millet production areas.Given the consideration of that,millet samples from three main producing areas across China are taken as the research objects and a hyperspectral-based traceability model of millet origin is proposed.The specific research content is as follows:(1)A hyperspectral-based traceability model of millet origin is proposed:SNV-SCARS-PSO-SVM.Adopting eleven preprocessing methods such as multivariate scattering correction(MSC)and standard normal transformation(SNV)to reduce noise reduction and filter.The best preprocess method is analyzed as SNV based on based on the preprocessed data;to reduce the dimensionality of the preprocessed data through continuous projection algorithm SPA),Stability Competitive Adaptive Re-Weighted Sampling Algorithm(SCARS)and other 4dimensionality reduction methods.The origin traceability model of millet is constructed based on4 modeling methods of partial least squares discriminant analysis(PLS-DA),random forest(RF),support vector machine(SVM)and particle swarm parameter optimization support vector machine(PSO-SVM).The best dimensionality reduction method and modeling method are determined as SCARS and PSO-SVM respectively according to the judgment effect of model traceability.Finally,the optimal combination model is SNV-SCARS-PSO-SVM.(2)The optimization of millet origin traceability model is realized by introducing parameter optimization algorithm.The particle swarm optimization(PSO)is introduced to optimize the process of optimizing the important parameters of the SVM model.Compared with judgement effect of different traceability models,the SVM model optimized by the PSO algorithm has the best effect on the traceability discrimination of millet origin among them.In contrast to the SVM model with random parameters,the accuracy and Kappa coefficients of SVM model incr eased from 84.25% and 0.766 to 92.86% and 0.900 respectively.The discriminant effect of PLS-DA model is second only to PSO-SVM,and the accuracy and Kappa coefficient are 92.63% and 0.893.The judging effect of the RF model is poor,and the judging effect of the SVM traceability model with random parameters is the worst.(3)The secondary traceability strategy of millet "producing area-producing base" is put forward to realize the improvement of millet origin traceability model.On the basis of the clustering phenomenon and interference effects discovered during the experiment,a secondary traceability strategy of “producing area-producing base” is proposed.The producing area of millet is firstly identified,and then to trace the producing base of millet.Given the consideration of the strategy to establish the traceability model of millet origin,the PSO-SVM model has obtained the best discrimination effect.The accuracy and Kappa coefficient of the model for the production area(central,northeast,and northwest)are 98.07% and 0.976,and the accuracy rate and Kappa coefficient are 97.50% and 0.950 in central area(Henan,Shanxi)).The accuracy and Kappa coefficient in the northeast regions(Heilongjiang and Inner Mongolia)are 97.44% and 0.949.The accuracy rate and Kappa coefficient in the northwest regions(Gansu,Ningxia,Shaanxi)are discriminated as 96.74% and 0.931.The accuracy and Kappa coefficient of the millet origin traceability model based on the "producing area-producing base" secondary traceability strategy has been significantly improved.It can be concluded that this study provides a high-efficiency and high-precision qualitative analysis model to apply the hyperspectral technology into the origin traceability of millet,provides a reliable technical support for the origin traceability of millet and other agricultural products,and provides a new research idea for millet and other agricultural products to identify the geographic information.