Discrimination of Absence or Presence of Pesticide Residue in Mulberry Leaf Using VIS-NIR Hyperspectral Imaging and Plsda

Abstract

Fast and accurate discrimination of pesticide residue in mulberry leaf is very important for sericulture and silk textile industry. Therefore, a hyperspectral imaging approach with the spectral range of 390-1050 nm was used for the discrimination of pesticide residue in mulberry leaf. 120 mulberry leaves samples including 60 samples without pesticide residue and 60 samples with pesticide residue were imaged by the VIS-NIR hyperspectral imaging system. ENVI software was used to explore the region of interest and extract the corresponding spectral data. Partial least square discriminant analysis (PLSDA) was used to establish the discriminative model for the discrimination of pesticide residue in mulberry leaf and the model achieved 98.33% calibration accuracy and 93.33% prediction accuracy. A total of 9 important wavelengths were selected according to the regression coefficient in the PLSDA model. A simplified PLSDA model was developed based on the important wavelengths and it achieved similar results (96.67% and 93.33%). The results showed that the model based on the selected important wavelengths was comparable to the model based on the full wavelengths, and it was feasible to use hyperspectral imaging technology for discrimination of pesticide residue in mulberry leaf.