Precise Detection And Visualization Of Pesticide Residues In Mulberry Leaves Based On Hyperspectral Imaging Information

Pesticide residues can be transmitted to people and animals through the food chain, which pose a serious threat to the health of them. Therefore, the most important thing is to seek an efficient, accurate and non-destructive method of detecting pesticide residues. At present, the detection of pesticide residues mainly depends on the chemical methods such as chromatography analysis and chromatography-mass spectrometry. Although the scholars have made the detection of pesticide residues in agricultural products using spectral techniques, the results of detection is not very ideal. In this paper, the hyperspectral imaging technology was applied to quantitative detection of pesticide residues in mulberry leaves. In this study, 71-1 mulberry leaves were used as the research object. Firstly, the effect of chlorpyrifos on the microstructure of mulberry leaves was studied by scanning electron microscopy and transmission electron microscope technology. The chemical amount of pesticide residues in mulberry leaves were determined by gas chromatography. Moreover, the mulberry leaves containing pesticide residues were carried out the analysis by hyperspectral imaging technology combined with the processing methods of spectra and image. The research contents are as follows:(1) Scanning electron microscope and transmission electron microscope were used to observe the microstructures of mulberry leaves sprayed with different concentrations of chlorpyrifos. It was found that opening degree and density of stoma in mulberry leaves treated by low concentrations of chlorpyrifos became larger, and the amount of osmiophilic particles increased. Besides, the mulberry leaves were treated by high concentrations of chlorpyrifos, the stoma of mulberry leaf was smaller,the mulberry leaf was thickened, osmiophilic particle was larger and increased in quantity.(2) In order to obtain more effective information, the spectrum and texture were combined by the information fusion. First of all, the band ratio algorithm and threshold segmentation method were used to image segmentation, and the whole mulberry leaf was set as region of interest(ROI), whose mean value of spectrum wasregarded as spectral data. Then the principal component analysis(PCA) was used to extract characteristic image of mulberry leaf. Textural features based on gray level co-occurrence matrix of characteristic image, include contrast, correlation, energy,homogeneity, entropy and mean.(3) SNV_detrending was selected to pretreat the spectrum of ROI to reduce the affection such as spectral noise, scattering spectra and baseline drift. And the successive projection algorithm(SPA), stepwise regression(SR) and weight regression coefficient(Bw) method were used to select characteristic wavelengths,which were respectively two characteristic wavelengths(461nm, 902nm), seven characteristic wavelengths(452nm, 453 nm, 527 nm, 602 nm, 814 nm, 957 nm, 982nm)and five characteristic wavelengths(450nm, 552 nm, 683 nm, 729 nm, 819nm).(4) Multiple linear regression(MLR), partial least squares regression(PLSR),least squares support vector regression algorithm(LS-SVM) and least squares-support vector machine based on bacterial colony chemotaxis optimization regression algorithm(BCC-LS-SVM) were used to establish models. These models were established based on spectrum or the fusion information of texture and spectrum on full bands or characteristic wavelengths to detect pesticide residues in mulberry leaves.The results show that the model established on fusion information of texture and spectrum on characteristic wavelengths has advantages. Among them, the Bw_BCC-LS-SVM regression model has the best performance, Rc and RMSEC were respectively 1 and 14.638, Rp and RMSEP were respectively 0.999 and 17.298.(5) The linear equation between pesticide residues and spectrum on characteristic wavelengths selected by SPA was1 2Y ?790.436-1435.91X-1396.59 X. Spectrums of all pixel in the hyperspectral image of mulberry leaf were introduced into the linear equation. Finally the false color images of pesticide residues in mulberry leaves based on hyperspectral image technology were obtained.