Non-destructive The Pesticide Residue Of Red Jujube Surface By Hyperspectral Imaging Technology

In the agricultural production,Chemical pesticide has the characteristics of convenient, economic and efficient on the controling plant diseases and insect pests.A large number of pesticides are used to control plant diseases and insect pests to improve agricultural production. Farmers’s means are fuzzy of spraying pesticide drug withdrawal period and safety harvest period.Makeing the fruit and vegetable surface and within have a large number of pesticide residues.If we eat the food of containing residual pesticides for a long time,it will seriously affect the health of human body. It can bring the potential harm to future generations. The food safety problems of pesticide residues has aroused the close attention of governments around the world and the organization. Traditional detection methods of pesticide residue have sample pretreatment process trival, reagent consumption, time consuming and other shortcomings.It has been difficult to meet the needs of nowadays pesticide residue.so looking for a rapid and nondestructive detection method have great significance.Taking LingWu long jujube as the research object,using near infrared hyperspectral imaging technology (900-1700nm) and chemometrics methods to distinguish the types of pesticide residues、 chlorpyrifos pesticide qualitative and quantitative nondestructive testing and quantitative detection of the imidacloprid pesticide residues on the surface of a long LingWu jujube.expected to find a method of rapid detection of pesticide residues, and providing the theory basis for LingWu jujube surface pesticide residue on-line nondestructive testing. The results of main research content as follows:(1)Based on the hyperspectral imaging technique to discriminate pesticide residues in the surface of jujube. By comparing the effects of different spectral pretreatment modeling, spectral convolution after preferred Savitzky-Golay smoothing establish partial least squares regression model (PLSR). Selected11characteristic wavelengths (957nm、1046nm、1103nm、1154nm、1219nm、1425nm、1502nm、1586nm、1607nm、1636nm) based on weighted regression coefficients of PLSR model.Established the linear discriminant model after the K-M conversion of11characteristic wavelengths,the recognition rate of chlorpyrifos and imidacloprid were92.3%; the recognition rate of Pyridaben and distilled water were100%; the overall recognition rate of correction model was96.43%.Using this model to predict and identify, predicted recognition results were91.7%.(2) Qualitative and quantitative detection the chlorpyrifos content of red jujube by hyperspectral imaging. Applying the linear discriminant model to discriminate different concentrations of chlorpyrifos pesticide on the red jujube surface.The discriminant calibration model accuracy is72.68%, but the prediction accuracy is only50%. On this basis, we studied the quantitative analysis of chlorpyrifos in the red jujube surface by the hyperspectral imaging. Established the PLSR model under full wave and characteristic wavelengths.By comparing the difference of the correlation and root mean square error, we optimized the model under characteristic wavelengths.The correlation and root mean square error of correction model is0.875,0.0025; The correlation and root mean square error of prediction model is0.816,0.0031. (3) Quantitatively detection the contents of imidacloprid on the red jujube surface by hyperspectral imaging. Respectively based on the raw spectral were pretreated by Savitzky-Golay convolution smoothing, Kubelka Munk value transformation and Kubelka Munk+Savitzky-Golay.We optimizing the Kubelka Munk+Savitzky-Golay was the best pretreatment by comparison the different spectral preprocessing. Established the PLSR model after the Kubelka Munk+Savitzky-Golay spectrum pretreatment. Based on the PLSR model weighted regression coefficient selected six characteristic wavelengths (990nm、1022nm、1270nm、1404nm.1583nm、1639nm). Established the PLSR model and support vector machine regression (SVR) model under whole band and characteristic wavelengths. Optimizing the PLSR model was the best model by comparison. Compared the PLSR model effect under the whole band and characteristic wavelength.the results show that: The correlation coefficient, the root mean square error of the PLSR model under characteristic wavelengths were0.86,0.0003; correlation coefficient and the root mean square error of Prediction model were0.85,0.0003. based on the analysis of all aspects of the factors, established model under characteristics wavelength is superior to the full band model.