Identification of Different Concentrations Pesticide Residues of Dimethoate on Spinach Leaves by Hyperspectral Image Technology

Abstract

Taking spinach leaves with different concentrations of dimethoate as the research object, the feasibility of identifying spinach leaves containing different concentrations of dimethoate pesticide residues based on hyperspectral imaging and machine learning algorithms was discussed. The band range of hyperspectral images between 900 to 1700mm were scanned by hyperspectral imaging system. The region of interest (ROIs) of the leaves were selected by ENVI and corrected by multiple scatter correction (MSC). The principal component analysis (PCA) was used to analyze the spectral data, and the results showed that the PCA can effectively discriminate spinach samples with different concentrations at the visual level. In addition, the chi-square test feature selection algorithm was combined with support vectors classification (SVC), K nearest neighbor (KNN), random forest algorithm (RF), and linear discriminant analysis (LDA) respectively. The average and standard deviation of the prediction accuracy of the 10-fold cross-validation was chosen as evaluation methods. By comparison, chi-square test combined with LDA was the optimal model and the selected characteristic wavelengths were 1445.8, 1449, 1452.3, 1455.5, 1458.7, 1462, 1465.2 and 1468.4nm. The prediction accuracy and standard deviation of the model were 0.997, 0.008. The results showed that spinach leaves containing different concentrations of dimethoate pesticide residues could be accurately identified based on hyperspectral imaging.