Research On Nondestructive Detection Of Nanjiang Jujubes Internal Quality Based On Near-infrared Hyperspectral Imaging Technique

Red jujube in Xinjiang,world-famous,due to the special geography,climate,environment,Southern Xinjiang red jujube made its surface morphology,or internal quality,obtained the widespread approval.The size of the surface morphology,plumpness can be recognition to the eye,but the measure of its internal quality is not easy.Traditional detection method of red dates internal quality is not only time-consuming,but also destructive or intrusion,and waste liquid and waste residue of testing process damage the environment.The rapid,efficient,environmental friendly detection means is imminent.Near infrared spectrum technology overcomes the drawback of most of the above,but its point detection and visual defects can not satisfy the requirement detection of internal quality of red jujube.Hyperspectral image technology arises at the historic moment,it not only contains the advantages of near infrared spectroscopy,and is a unity of area detection,map,high resolution,can be achieved internal physical and chemical composition and surface defect inspection.,The paper describes the pre-modeling work from the experimental materials and the physical and chemical analysis methods,the other functions of high spectra and the method of stoichiometry.The main research content is as follows:(1)Southern Xinjiang red jujube best spectral data preprocessing methods,the pretreatment methods of article study include derivative method,the multiple scattering correction,standard normal variable transformation,the effects of vector normalization of the modeling effects.Through the above pretreatment method to establish the model of total sugar and water partial least squares(PLS)and BP neural network(BP ANN)of nanjiang white period jun jujube,the conclusion is the multiple scattering correction pretreatment is superior to other methods.(2)The choice of the characteristics wavelength variables of the Xinjiang red jujube,the study of the characteristics wavelength variable methods including correlation coefficient method,the continuous projection method,genetic algorithm and the competitive adaptive weighted algorithm.Research shows that: the characteristics of genetic algorithm to choose the total sugar wavelength variables prediction performance is better than other methods,characteristic wavelength of 942 nm,1 069 nm,1 110 nm,1 179 nm,1 284 nm,1 494 nm,1 570 nm,1 626 nm,1 661 nm,1665 nm,1 672 nm,1 693 nm.Through a competitive adaptive weight weighted algorithm choose the moisture characteristic wavelength variables prediction performance is better than other methods,characteristics of wavelength of 957 nm,989 nm,1 085 nm,1 107 nm,1 179 nm,1 195 nm,1 271 nm,1 419 nm,1 439 nm,1 511 nm,1 529 nm,1 598 nm,1 661 nm.(3)The study of red jujube of Southern Xinjiang modeling method,the choice of pretreatment method using multiple scattering correction,total sugar and water characteristic wavelength variable selection of jun jujube using separately genetic algorithm and competitive adaptive weighting algorithm,and using partial least squares regression analysis and BP neural network for modeling.Analysis shows that: in the established model of total sugar and water of the jun jujube,the prediction model of BP neural network are better than model by partial least squares.The jun jujube total sugar and water model about correlation coefficient(Rp)respective reach 0.918 4 and 0.889 1.Total sugar and water model of jun jujube about the prediction to standard deviation(RMSEP)respective reach 0.931 0 and 2.198 0.