The Nondestructive Inspection Research Of Pesticide Residues Contamination On Fruit Surface Based On Hyperspectral Imaging And Near Infrared Spectroscopy Technology

The research object of this paper is gannan navel orange. Hyperspectral imaging technology and near infrared spectroscopy are applied on gannan navel oranges to nondestructive inspection one of its safety qualities——residue of pesticide. Non-destructive testing has the characteristics such as rapid, simple, online testing and so on,so the study of pesticide residue using the method of nondestructive inspection has practical value.The cultivated area of navel orange in Jiangxi province ganzhou city is the first in the world, which annual output is the third in the world and ganzhou city has the largest navel orange producing area in our country. Gannan navel orange is one of China's major export fruit, food safety quality is the main measure quality indicators, therefore the detection of surface pesticide residues on navel oranges had an important significance.A hyperspectral imaging platform is established based on hyperspectral imaging technique for this research.1:20,1:100 and 1:1000 pesticide solution are prepared,and then three different concentrations solution are dropped to the orange surface with three different amount.Then put the fruits in well-ventilated and shady place, and take the images 168 hours later. And then hyperspectral images of gannan navel oranges are collected from wavelength 400nm to 1000nm. The wavelength range from 625nm to 725nm is selected for testing.Imaging calibration, image denoising, median filtering and region of interest these methods are used to process gannan navel oranges'hyperspectral images. Principal component analysis (PCA) is used to analyze the images, and obtain the images on feature wavelengths. Proper image processing method and PC-3 image are used to test the pesticide of navel oranges'surface.Secondly, in this paper laser induced fluorescence hyperspectral image technology is used to acquire the fluorescence of navel oranges surface, and the average of the fluorescence as the basis of data analysis. Partial least squares regression (PLS) method is applied to establish forecasting model and the best combined spectral band are found. SVM method is used on the basic of best spectral bands for establishing the forecast model of dichlorvos pesticide residues. Four Spectral region (453-501.8nm,604.4-650nm,704.3-750.6 and 754.8-801.5nm) are selected in effective spectrum regional scope(from453nm to 801.5nm). The final pesticide residue forecast model are established by SVM. The correlation coefficient between actual group and predicted group is 0.801. The Research indicates that the detection of pesticides residues dichlorvos on the surface of the navel oranges by using laser induced fluorescence hyperspectral image technology is feasible, and it will provide theoretical basis for realizing online nondestructive testing of pesticide residues on agricultural products. In addition, near infrared spectral technology is applied for inspecting pesticide residues gannan navel oranges surface in this paper.Firstly, the paper qualitative analysis the pesticide,which is on the fruit surface. the main use of directly on the fruit surface collection of near infrared spectral, use genetic algorithm (Genetic Algorithm, GA) for optimization of the entire spectral region, determine the best feature spectrum interval and wavelength, establish a predicting model, which is for unknown samples to pesticide contamination and whether navel oranges by the identified what pesticide contamination with forecast.The results showed that:the best iteration number is 40, and ultimately determine the number of input variables is 71, and then point the recognition accuracy rate of 99.57%. Application method of GA-SVM for input variables of the model for further screening, get features wavelength is 71, the model's predictive power has not changed, Indicating that application of GA-SVM method is better than full spectrum model which is more stable and concise.Based on the qualitative detection, this paper quantitative analysis of the fruit surface pesticide.Through collecting directly on the fruit surface of near infrared spectral of spectrum, use four different pre-treatment to pretreatment the spectrum, application of partial least squares and least squares support vector machine method for the treatment of spectral data modeling and predicting.Quantitative analysis to determine the best surface navel oranges pesticides residues spectral pretreatment, quantitative analysis to determine the best surface navel oranges pesticides residues spectral pretreat methods.The results are as follows:(1) For detecting Mancozeb pesticide residues, considering the number of PC and the correlation coefficient,the best methord of preprocessing is the first derivative. The correlation coefficient R of calibration group and prediction group is 0.8587and 0.7719 respectively, RMSEP of calibration group and prediction group is0.3309 and 0.3343 respectively.(2) For detecting dichlorvos pesticide residues, considering the number of PC and the correlation coefficient,the best methord of preprocessing is the first derivative.The correlation coefficient R of calibration group and prediction group is 0.9895 and 0.8174 respectively, RMSEP of calibration group and prediction group is 1.5729 and 6.2598 respectively.