Quality Detection And Platform Development Of Dried Jujube Based On Hyperspectral Imaging Technology And Machine Learning

Jujube contain very high vitamins and are loved by people.Fresh jujube has rich nutrition and unique flavor,but it is very perishable and has a short shelf life.Through the use of dry jujube storage technology,the supply period and transportation distance of jujube can be extended,thus improving the value of jujube and ensuring the income of jujube farmers.During the storage of dried jujube,the quality of water,total acid and total sugar have an important impact on the taste and flavor of jujube during storage,which greatly affects the economic benefits of jujube farmers.Through hyperspectral imaging technology,the quality of agricultural products can be quickly and accurately detected,and the physiological changes of water,total acid and total sugar of dried jujube during storage can be studied to determine the appropriate water content,thus delaying the decline of product quality,and providing an important basis for agricultural production.The research focus of this thesis includes:(1)Identification of different storage periods of dried jujube based on hyperspectral imaging technologyAfter measuring the moisture,total sugar and total acid of the dried jujube samples,the Monte Carlo algorithm is used to remove the singular value,and then the content gradient method is used to divide the training set,verification set and test set,and the spectral data analysis is pretreated with SG and SNV,so as to build partial least squares regression(PLSR),support vector return(SVR),RF,LeNet and ResNet models,and establish the water Prediction model of internal quality attributes of jujube based on total acid and total sugar.The results show that the overall prediction model performs well,with an average decision coefficient greater than 0.70 and an average RPD greater than 1.4.However,the prediction model based on NIR is better than Vis-NIR.Finally,according to the NIR data,the prediction model under multiple storage periods is established.The results show that the performance under multiple storage periods is better than that under single storage period,and the deep learning shows stable and excellent results.(2)Detection of moisture,sugar and acid content of dried jujube based on hyperspectral imaging technology and machine learningObtain hyperspectral data of dried jujube samples and measure the corresponding moisture,total sugar and total acid.After the singular value is removed by Monte Carlo algorithm,the content gradient method is used to divide the training set,verification set and test set,and the spectral data analysis is pre-processed by SG and MSC,so as to construct partial least squares regression(PLSR),support vector return(SVR),RF,LeNet and ResNet models,which are used to predict the internal quality attributes of jujube under a single storage period.The results show that the overall prediction model performs well,with an average decision coefficient greater than 0.70 and an average RPD greater than 1.4.However,the prediction model based on NIR is better than Vis-NIR,and then based on NIR,the prediction model under multiple storage periods is established.The results show that the performance under multiple storage periods is better than that under single storage period,and the deep learning shows stable and excellent results.(3)Development and design of the system for quality detection of dried jujubeBased on the identification of different storage periods of dried red jujube and the detection requirements of water content,total acid and total sugar content,a detection system of internal quality attributes of red jujube was designed.The main function modules of the system design are: user registration and login,identification of jujube storage period,prediction of jujube internal moisture,sugar and acid content.The system provides convenience for the operators of jujube processing plants to inspect the quality.