Quality Detection And Research Of Yongquantangerine Based On Near Infrared Spectroscopy Technology

With the development of social economy,people's living conditions have been greatly improved,and higher requirements are put forward on the quality of consumer fruits and vegetables.At present,the main methods of fruit and vegetable quality testing are sensory evaluation based on subjective experience and destructive physical and chemical testing methods.The test results of the former cannot quantify fruit quality information,while the latter cannot be tested in a large-throughput and automated manner.It also causes a certain amount of fruit and vegetable loss and cannot be sold again.The work takes the Zhejiang Yongquan tangerines near the harvest period as the research object.Different batches of tangerines are continuously picked,and the near-infrared spectroscopy and the Soluble solid content data of the tangerines are collected successively.SSC is an important indicator of tangerines quality.The spectral data combined with chemometrics is used to model and analyze the quality of tangerines.However,there are certain differences in the quality of tangerines in different picking periods,which are reflected in the spectra of different batches of tangerines,which leads to large deviations in the predictions of different batches of tangerine models.To predict the quality of different batches of tangerines,it is necessary to transfer between different prediction models in reducing model errors.In addition,the prediction of the shelf life of postharvest tangerines based on spectral analysis technology is beneficial to the study of the quality of different batches of tangerines and the classification of freshness during different storage periods.The main research contents and conclusions are as follows:(1)Seven batches of Yongquan tangerines are collected continuously,and the picking interval was one week.Seven batches of tangerines are measured by two portable near-infrared spectrometers(Beijing Yishida Quality Selection Company's H100 F and Shandong Haineng Instrument Company's N100),with collecting near infrared spectrum information and measuring the corresponding SSC.Then the Principal component analysis and Mahalanobis distance algorithm are used to check the spectrum of each batch of samples and eliminate abnormal samples and index abnormal samples.Finally,a different number of adjacent batches(three batches,two batches and one batch adjacent to the prediction set)are used as the training set to establish a Partial least square(Partial least square,PLS)correction model to predict the next batch for the second sample,the result shows that a batch of oranges in the nearest prediction set has a better prediction effect as the training set.(2)For tangerines from different picking periods,the corresponding internal quality(mainly SSC)of tangerines has a certain difference,which is reflected in its spectral information,so that there is a systematic error in the prediction of different batches of tangerine spectra.In this paper,a batch of tangerines in the nearest prediction set is used as the training set,and a variety of spectral preprocessing methods are compared.The Domain adaptive partial least square regression algorithm is used to correct the spectral matrix between different batches.The comparison finds that the spectral modeling effect without any preprocessing is considered to be the best;compared with the PLS model,the Root mean squared error of prediction of the DA-PLS model is at least 3.20 %lower than that of the PLS model,DA-PLS correlation coefficient of prediction is 3.2% higher than that of PLS.It is concluded that after the correction of the spectral matrix,the prediction performance of the DA-PLS model is better than that of the PLS,and it has a lower prediction error.(3)According to the shelf-life quality of tangerines,the storage time of tangerines is designed to be up to 19 days.The change characteristics of the skin and internal components of tangerines during the shelf life are divided into three freshness levels.Then,the Soft independent modeling class analog and Partial least square regression discrimination analysis are used to classify the freshness of different storage periods.The results show that the discriminant analysis effect of PLS-DA prediction model is better,and the prediction accuracy is higher.Among them,the discrimination accuracy of the correction model in PLS-DA reaches 99.39%,while the maximum discrimination accuracy of the correction model built in the training set corresponding to the SIMCA model is only 83.75%,and the maximum accuracy of the prediction set is 77.50%.