Study On Spectrum Analysis Method Of Vis-NIR Based On Pesticide Residues On The Surface Of Leek

With the development of economy in China, the quality of people’s life has been improved a lot. Once upon a time, people eat food just for full and now people also care for food nutrition. It proved that the Chinese’s demand for food are higher and higher. So it is important to pay attention to the safety and quality of food. The problem of pesticide residues on agricultural products has become an obstacle to the ensurance of food safety. During the growth process of agricultural products, some illegal traders use pesticide for their own interests in an incorrect way, which results in the pesticide residues of agricultural products exceeding the standard. Thus threatens the health of people. Therefore, there is an urgent need to find a method to detect pesticide residues in a fast and economic way. The visible-near-infrared spectra method has become a kind of method for fast analysis and detection, which has been applied to the detection of pesticide residues. This paper took leek as the experimental object and used the visible-near-infrared spectra method for a nondestructive testing research.This paper used ASD spectrometer to detect leek samples with chlorpyrifos to receive the spectral data. The wavelength of spectrum was in the range of 350-2500 nm and the 350-419 nm band with more noise has been removed. Then the spectral data was to be preprocessed. The preprocessing method was First Differential(FD), Second Differential(SD), Multiplicative Scatter Correction(MSC) and Standard Normal Variate(SNV). The quantitative analysis models used to detect leek samples with chlorpyrifos were established after the preprocessing using Partial Least Squares(PLS) and the Boosting. The prediction results showed that the optimal models of PLS and Boosting were all used by the First Differential to preprocess and the result of Boosting is better. By computing the correlation between the hyperspectral characteristics parameters and the amount of pesticide residues, this paper chose spectral data of 490-560 nm as input variables of PLS and Boosting. The calculation results showed that compared with the regression model of full-spectrum data, the precision of the model selected by characteristic wavelengths was slightly lower, but the difference is small. Considering that the complexity of calculation and the time required, the model which used characteristic wavelengths as input variables was better than the model of full-spectrum and the prediction results of Boosting is better than that of PLS.This paper used spectral data acquired by ASD spectrometer to distinguish the types of the pesticide residues on the surface of leek. Firstly, full-spectrum data of 350-2500 nm was preprocessed. The preprocessing method was First Differential(FD) and Second Differential(SD). After removing the 350-419 nm bands with more noise, the spectral data which had been preprocessed was used as input variables to establish disaggregated model of SVM and Boosting. Through the classification accuracy of the experiment we could see that, in the SVM model, the classification accuracy of calibration set and prediction set of the data after Frist Differential was 1 and 0.9048. It was significantly higher than that of the original spectrum and the Second Differential. In the Boosting model, the classification accuracy of calibration set and prediction set of the data after Frist Differential was 1. The results of experiment showed that the data after First Differential behaved the best in the SVM and Boosting. It could identify the types of the pesticide residues on the surface of leek.