Research On Rapid Detection Method Of Rice Protein Content Based On Raman-NIR Spectral Fusion Technolog

Rice is one of the most important food in the world.It has the characteristics of wide planting,balanced nutrition and large edible population.As an important part of human cells and tissues and a participant in life activities,protein is an indispensable nutrient for the human body.Rice contains about 7%to 8%protein.Rice protein is a plant protein with excellent quality.It has high amino acid composition and high nutritional value.The protein content also affects the eating quality of rice.High protein rice is usually hard and easy to deteriorate during storage.The rice is also light yellow,resulting in a decrease in the appearance and eating quality of rice.Therefore,the amount of protein content is an important evaluation index of rice quality.However,the current traditional chemical protein detection methods are time-consuming and laborious.How to use spectroscopy to achieve rapid detection of protein content has become a research hotspot in recent years.Therefore,this paper attempts to use data fusion technology combined with spectral characteristic wavelength optimization to quickly detect rice protein content based on Raman spectroscopy and near infrared spectroscopy(NIR)spectroscopy.In this paper,150 rice samples were collected from 5 regions of Chahayang,Wuchang,Fangzheng,Xiangshui and Jiansanjiang in Heilongjiang Province,30 samples were collected in each region,and a total of 150 rice samples were taken as the research objects.Raman and NIR spectral data were obtained by scanning.The spectral data dimension reduction and data fusion technology of principal component analysis(PCA)combined with spectral characteristic wavelength optimization algorithm were used to establish partial least squares(PLS)regression prediction model for rice protein content detection.The main research contents of this paper are as follows:(1)Based on the idea of particle swarm optimization(PSO),an improved binary particle swarm optimization(IBPSO)algorithm for spectral characteristic wavelength optimization is proposed.In order to evaluate the characteristic wavelength selection performance of IBPSO,a variety of characteristic wavelength optimization algorithms including IBPSO are combined with data fusion technology to optimize the spectral characteristic wavelengths,and a PLS regression prediction model is established.The performance and prediction ability of PLS models based on spectral data optimized by different algorithms were compared.The results show that when the data layer fusion strategy is applied,the prediction determination coefficient(R_p~2),root mean square error of prediction(RMSEP)and mean relative error of prediction(MREP)of the model based on IBPSO for feature wavelength optimization are better than the other four classical algorithms.The results show that IBPSO can achieve efficient acquisition of high correlation modeling wavelength variables through the guiding optimization of particle value as’1’binary bits.(2)Based on the data layer fusion strategy of data fusion technology,this paper constructs Raman combined with near infrared(Raman-NIR)fusion spectrum to establish PLS regression prediction model for rapid detection of rice protein content.Several models were established to compare the performance of the models,which were single spectral full spectrum PLS model,Raman-NIR data layer fusion spectral full spectrum PLS model and Raman-NIR data layer fusion spectrum optimized by characteristic wavelength PLS model.The modeling results show that the PLS model based on the characteristic wavelength optimization of the Raman-NIR data layer fusion spectrum has the best performance,and the PLS model established by using IBPSO to optimize the spectral characteristic wavelength is better than the Raman-NIR data layer fusion spectrum full spectrum PLS model and the single spectrum full spectrum PLS model.The results show that the data layer fusion strategy that combines all the spectral information of the two can improve the prediction ability and stability of the model.On the other hand,it also confirms the complementarity between Raman spectroscopy and NIR spectroscopy.(3)In this paper,multiple PLS regression prediction models are established based on the feature layer fusion strategy of data fusion technology.They are Raman-NIR feature layer fusion PLS model after PCA dimension reduction and Raman-NIR feature layer fusion spectral PLS model after feature variables are extracted by various spectral feature wavelength optimization algorithms.The modeling results show that the prediction performance of the Raman-NIR feature layer fusion spectral PLS model after the feature variables are extracted by various spectral feature wavelength optimization algorithms is better than that of the Raman-NIR feature layer fusion PLS model after PCA dimensionality reduction.The Raman-NIR spectrum selected by GA-IBPSO characteristic wavelength selection algorithm has the best model.The results show that the high-level data fusion strategy can reduce the difficulty of model establishment and improve the prediction ability of the model.At the same time,it is confirmed that the combination of IBPSO and spectral data fusion technology can realize the rapid detection of rice protein content,which provides theoretical support for the development of related online detection equipment.