| Fresh jujube originated in China, owing to its strong adaptability of soil and climate, it had been widely cultivated. Fresh jujube is fond by people because of its bright color and sweet flavor. In order to ensure the quality of fresh Chinese jujube, reduce postharvest damage and increase the added value of postpartum, researching on the quality of fresh jujube detection with nondestructive detection technology has very important practical significance. In this study, fresh jujube was chosen as the research object. The external defects, the internal quality and the comprehensive quality of fresh jujube had been detected based on NIR spectroscopy, hyperspectral imaging technology and the multi-level information fusion technology. The results and conclusions of this study were as follows:(1) The different ripeness of jujube had been detected from the spectral information, texture information, color information and the fusion information based on hyperspectral imaging technology. The classification model of partial least squares (PLS) and least square support vector machine (LS-SVM) were established in texture features (contrast, correlation, energy, homogeneity, variance, mean and entropy), color features (RGB model, HSV model and YCbCr model) and characteristic wavelengths, which were selected by two-dimensional correlation spectroscopy(TDCS), correlation coefficient (RC) by PLS model, successive projection algorithm(SPA) and competitive adaptive reweighted sampling (CARS), respectively. The results indicated, with the exception of model built by characteristic wavelengths, the discriminant accuracy was useless of discrimination model built by texture and color features. Therefore, the multi-sensor information fusion technology which fused spectral, texture and color information had been used to discrimination the ripeness of fresh jujube. After comparing the model, the LS-SVM model was built by characteristic wavelengths which selected by CARS algorithm, texture features without mean index and HSV color model, had best optimal prediction accuracy, the prediction set of discriminant accuracy reached 96.67%.(2) The internal quality (moisture content, soluble solid content, vitamin C content, protein content and firmness) of fresh jujube had been detected based on NIR spectroscopy. In order to investigate prediction the performance of the quality index in fresh jujube comprehensive, the prediction models were built by moisture content, soluble solid content, vitamin C content, protein content and firmness using NIR spectroscopy. The Influence of the accuracy of the model was analyzed by a variety of transformations and pretreatments of spectral data. The optimal characteristic wavelengths method of the quality index of fresh jujube was analyzed by lots of algorithms, which provided data and theory support for the research and development of an online detection device.(3) In view of the difference of moisture content has a very important influence on the near infrared spectrum, water compensation prediction models about SSC, VC, protein content and firmness proposed. On the basis of the characteristic wavelength of each extracted quality index, the water compensation model of the index was established by fusion water compensation wavelengths (removed the superimposed or be close to water characteristic wavelengths) and water content. The results showed that the prediction accuracy of water compensation models was decreased for each quality. Hence, there was a certain correlation relationship between the water content and the other quality indexes, and then the correlation relationship between the quality indexes was analyzed. The results indicated that the water content had an extremely significant correlation relationship with SSC, VC content and protein content in the 0.01 levels and had a corresponding research conclusion in the accuracy of the model of the water compensation.(4) An internal comprehensive quality index had been established by the 5 quality indicators. Factor analysis was carried out on 5 internal quality index of fresh jujube, in first two principal factors of the cumulative contribution rate had reached 82.738%, which means the main information of the original basic variables can be reflected by the first two factors. The modeling accuracy of the internal comprehensive quality index established in the different data rotation was compared and analyzed. Finally, the comprehensive quality index of non-numerical rotation was selected as the optimal internal comprehensive quality index, and the characteristic wavelength of the optimal index was extracted and analyzed. The results indicated that the CARS-LS-SVM model was the optimal prediction model of comprehensive quality of fresh jujube, the R and RMSEP of prediction set were 0.9241 and 6.0635, respectively.(5) Spectral information and texture information of the image data had been fusion based on hyperspectral imaging technology to detect the quality of internal and external of fresh jujube at the same time. For the first time, according to various defects were emerged before fresh jujube picked, a new conception was innovatively proposed, including shrink, crack, insect damage, black spot and peck injury. A natural damage (external quality) detection model had built from the spectral layer (characteristic wavelengths) and the image layer (operator of Sobel and region growing algorithm). At the same time, a prediction model of the characteristic wavelengths was established for prediction of the internal comprehensive quality (internal quality) of fresh jujube. And then, based on the information fusion technology, spectrum information and texture information extracted by hyperspectral imaging had been fused to detect natural damage and internal comprehensive quality at the same time, the discrimination accuracy of natural damage (external quality) reached 92.31%, the R and RMSEP of the prediction accuracy of comprehensive quality (internal quality) were 0.9538 and 8.1879, respectively.(6) The migration of the hyperspectral data and near infrared data was been analyzed to compare the differences of the performance and migratory aptitude of spectral data between near-infrared spectroscopy and hyperspectral imaging. The results indicated that the prediction accuracy of spectral data collected from hyperspectral imaging system was lower than spectral data collected from near infrared spectrometer, but the accuracy still can be used to predict the quality of natural damage and normal fresh jujube, which indicated that the two spectrometer acquisition of spectral data with good mobility. When the internal quality was studied, the near infrared spectrometer was more reasonable, basic on its small interval data, a wide range of wavelengths, when the external quality or internal and external comprehensive quality need to be sampled or measured, the hyperspectral imaging system was more reasonable.(7) The diffuse reflection on-line detection system was built, and the online detection model of the internal comprehensive quality of fresh jujube was established. In order to realize the on-line detection of the internal quality of fresh jujube, the on-line near infrared diffuse reflectance spectroscopy was built to detect the internal quality of fresh jujube. The hardware system of the device was consisted of an electromechanical control unit, a data acquisition unit and a data processing unit, the hardware system was consisted of data processing software. According to the actual needs of detection process, the optimal calibration CARS-LS-SVM model was integrated in diffuse reflectance spectroscopy on-line detection software. In order to verify the stability and reliability of the on-line detection system, related replication experiment had been conducted. The study confirmed that the NIR spectroscopy on-line detection system could be used to predict the internal comprehensive quality of fresh jujube. This research provides fresh ideas for rapid non-destructive internal comprehensive quality of fresh jujube. The results are promising for promotion of the non-destructive detection technology for practical, which is also of great significant in improving the level of detection in the fruit industry in China. |
PDF Full Text Request