Study On Non-destructive Detection Common Defects Of Lingwu’s Jujube Based On Hyperspectral Imaging Technique

LingWu long jujube, as one of characteristic agricultural products in Ningxia, is favored by the broad consumer for its high nutritional value. However, the external quality of long jujube must affect directly its sale and storage. In the traditional detection method, it has several disadvantages such as time-consuming, laborious and low efficiency, etc, Hyperspectral imaging technique has become an important trend to employ nondestructive testing of fruit quality, because it simultaneously has the merit of spectral technique and imaging technique.The paper, with jujube in Lingwu as the study object, collected spectral data by Vis-NIR and NIR hyperspectral imaging system respectively. On the one hand, the SVM and SIMCA method were applied to data analysis to establish the classification models of intact, crack, bruised and insect jujube. On the other hand, the imaging information of intact, crack, bruised and insect jujubes, combined the optimal wavelength PCA and band ratio algorithm with imaging processing, were contributed to establish recognition models of the common defects of jujube. Hence, the studied work would make preparations for the next step development of real-time, fast, nondestructive detection system online to provide theoretical basis. The main research results of the paper are as follows:Firstly, the influence element of collecting spectra through two set of hyperspectral imaging systems (spectral range400-1000nm and spectral range900-1700nm) were investigated, such as acquisition parameter settings, background and color. At the same time, the acquisition time for bruised jujube was determined to be after24h.Secondly, the model of performance established by the original spectral reflectance curve was better than that of pretreatment (MSC/SNV/FD) and spectral parameters (R/A/K-M).Thirdly, the PCA method was applied to reduce the spectral dimension of intact, insect, crack and bruised.jujubes through two hyperspectral imaging systems. The five optimal wavelengths(420、521、636、670、679nm) in Vis-NIR range were extracted and in NIR range the four optimal wavelengths(1028.1118、1359、1466nm) were obtained. The SVM and SIMCA model based on the full and optimal wavelengths had been compared. The result showed: the established SVM and SIMCA model based on full wavelengths can replace that of optimal wavelengths, the C-SVM type of SVM model is prior to the Nu-SVM type of SVM model, linear Kernel of SVM model superior the other Kernel of SVM model, the SVM and SIMCA model in NIR range is better than that of Vis-NIR range.Fourthly, imaging information of intact, insect, crack and bruised jujube in Vis-NIR range was used to make PCA and extracted the six optimal wavelengths(545、569.670、732.881、910nm) of intact jujube, the five optimal wavelengths(569.603.675.761.910mm) of insect jujube, the three optimal wavelengths(670.780.910nm) of crack jujube, the four optimal wavelengths(502.603.670.886nm) of bruised jujube to make PCA again. The identify result was100%,64%,46%,0, respectively. To improve identify ratio, the band ratio algorithm was applied to identify further. The results showed: the R910/R569obtained better identify ratio of insect jujube that develop to84%and the R910/R670made the identification ratio of crack jujube up to69%,but the identification ratio of bruised jujube was not obvious.Fifthly, imaging information of intact, insect, crack and bruised jujube in NIR range were used to make PCA and extracted the four optimal wavelengths(1028、1109、1312、1449nm) of intact jujubes, the four optimal wavelengths(1025、1109、1312、1449nm) of insect jujubes, the seven optimal wavelengths(103、1112、1225、1312、1392、1449、1461nm) of crack jujubes, the four optimal wavelengths(1034、1112、1312、1449nm) of bruised jujubes to make PCA again. The identify results were100%,72%,86%,100%, respectively. To improve identify ratio, the band ratio algorithm was applied to identify in future. The R1449/R1034could obtain better identify ratio of insect jujube that develop to90%and the identification ratio of crack jujubes was not obvious.In conclusion, through compared the two hyperspectral imaging systems, the result showed hyperspectral imaging system of NIR range was better than that of Vis-NIR range.