Establishment And Application Of Quality Model Of Fresh-cut Lettuce Based On Hyperspectral Technology

The freshness of lettuce has always been the core issue of the lettuce quality,because the metabolism of tissue cells is vigorous and the respiratory rate is very high,which leads to the wilting and aging of fresh-cut lettuce.The automatic processing of lettuce needs accurate and rapid quality classification technology of lettuce freshness at the same time.Therefore,it is urgent to study the monitoring model and evaluation method of lettuce freshness to provide theoretical basis for the rational development and utilization of lettuce resources.The main contents of this paper are as follows:(1)The fresh-cut lettuce sold in the market was taken as the research object,and the comprehensive evaluation method of lettuce freshness and quality was established by sensory evaluation in this study.First,based on the freshness of leafy vegetables and the sensory evaluation results of fresh-cut lettuce,the key indicators that mainly affect the freshness quality of fresh-cut lettuce were determined.The moisture content,soluble solids content,chlorophyll content,and L*,a*and b*values and MDA content.Correlation analysis was then performed on the sensory evaluation results of fresh-cut lettuce and the measured physical and chemical indicators.As a result,it was found that moisture content and soluble solid content were the main factors affecting the freshness of fresh-cut production,and their cumulative contribution rates to freshness reached 87.399%and 76.022%.Finally,based on the results of correlation analysis,a freshness quantification model of fresh-cut lettuce was constructed to achieve the purpose of quantifying the freshness quality of fresh-cut lettuce.(2)The spectral information of fresh-cut lettuce leaves was obtained by using hyperspectral technology in the 350-2500 nm band,and the spectral information was pre-processed.Including:data normalization,data smoothing,derivative method,multivariate scattering correction and standard normal transformation,so as to achieve the purpose of reducing spectral information noise and improving the accuracy of spectral information.Besides,the corresponding physical and chemical indicators were extracted by principal component analysis and continuous projection algorithm.Finally,four prediction models of multiple linear regression,partial least squares,artificial neural network and least squares support vector machine were constructed based on the characteristic spectrum information.(3)Based on the spectral information of fresh-cut lettuce leaves and corresponding physical and chemical values,a partial least squares-artificial neural network composite model for predicting the moisture content of fresh-cut lettuce was first constructed,and the correlation coefficient of the fresh-cut lettuce moisture content prediction model was constructed,R2=0.9653 and RMSE=0.134.The six characteristic wavelengths selected were:1323,1345,1382,1417,1794,and 1795nm.Then,a partial least squares-artificial neural network model was constructed to predict the soluble solids content of fresh-cut lettuce.The correlation coefficients of the prediction model were R2=0.9568 and RMSE=0.0017.The five characteristic wavelengths selected were:961,974,985,1120,and 1231 nm.Finally,the prediction model of chlorophyll content and malondialdehyde content in fresh-cut lettuce was constructed.Among them,the correlation coefficient of prediction model of fresh-cut lettuce chlorophyll content The six characteristic wavelengths selected were:477,569,625,712,739,and 741 nm.The composite model of freshness of fresh-cut lettuce was determined by compound processing different prediction models of physical and chemical indicators.The determination coefficient of the model=0.97,and the root mean square error=0.0103.(4)In the actual processing and sales process,fresh-cut lettuce is usually sold in plastic wrap.Considering the influence of fresh-cut lettuce film materials on the prediction model,the influence of three kinds of film materials:Polyethylene film,vinyl chloride film and polyvinylidene chloride film on the accuracy of the composite prediction model was determined.Based on the results of the discriminative confusion matrix of the least squares support vector machine(LS-SVM)model,the existence of cling film had little effect on the discrimination results of fresh-cut lettuce using hyperspectral technology.The correct recognition rate of the discrimination was 85.7%,the correct recognition rate of the vinyl chloride film was 88.4%,and the correct recognition rate of the polyvinylidene chloride film was 94.1%.The comparison found that although the three cling films all reduced the correct identification of the best model,the actual impact was small.The composite prediction model of lettuce freshness established in this study had strong robustness.This research was based on hyperspectral technology to study the freshness detection method of fresh-cut lettuce.Compared with traditional detection methods,it is non-destructive,fast,easy to operate,low cost and easy to popularize.Design and development provide the theoretical basis.