Identification of pesticide residue level in lettuce based on hyperspectra and chlorophyll fluorescence spectra

Abstract: Fast identification of pesticide residue level in lettuce leaves plays a key role in the test of food safety. In order to identify the different concentrations pesticide residues of lettuce leaves in a fast and nondestructive way, the hyperspectra coupled with chlorophyll fluorescence spectra was used in this research. Transmission electron microscopy (TEM) was used to identify the microstructure changes of lettuce leaves under different concentrations of dimethoate residue. Besides, a method involving wavelet transform and MD-MCCV algorithm (WT-MD-MCCV) was developed for identifying the optimal wavelengths of the spectral data. The hyperspectra and chlorophyll fluorescence spectra data of 150 lettuce leaf samples at five different concentrations of pesticide residues were obtained using hyperspectral data acquisition device and Cary Eclipse Fluorescence Spectrophotometer. The combination of Savitzky-Golay (SG) algorithm and SNV algorithm (SG-SNV) preprocessing algorithms was used to preprocess the raw spectra. In addition, Principal Component Analysis (PCA), Successive Projections Algorithm (SPA) and wavelet transform coupled to MD-MCCV algorithm (WT-MD-MCCV) were applied to identify the optimal wavelengths of raw spectra including hyperspectra data, chlorophyll fluorescence spectra data and hyperspectra coupled with chlorophyll fluorescence spectra data. Support vector regression (SVR) was applied to build the prediction models based on preprocessed spectra feature in characteristic wavelengths coupled with different spectral data. The results showed that with the increase of the concentration of dimethoate pesticide spraying, lettuce chloroplast number of osmiophilic particles increased and the starch granules decreased. Besides, the intercellular space of lettuce leaves increased gradually, with the increase of dimethoate concentration. Different concentrations of pesticide residues of lettuce in the near infrared and fluorescence spectrum have a certain difference. In addition, the related parameters of the three preferably prediction models were Rp2=0.956 and RMSEP=0.018, Rp2=0.937 and RMSEP=0.161, Rp2=0.987 and RMSEP =0.005, respectively, using WT-MD-MCCV algorithm combined with hyperspectra data, chlorophyll fluorescence spectra data and hyperspectra coupled to chlorophyll fluorescence spectra data. WT-MD-MCCV algorithm combined with hyperspectra and chlorophyll fluorescence spectra data performed best among the nine SVR models and the hyperspectra coupled with chlorophyll fluorescence spectra can be used to identify the pesticide residue level in lettuce leaves.