Tomato Pesticide Residue Detection Method Based On Hyperspectral Imaging

In this study, the possibility of non-destructive pesticide residues detection method on tomato was investigated using hyperspectral and low cost imaging technology and prediction classification models, comparing it with models based on hyperspectral images acquired by a commercial camera (Pika L). Spectral data from samples of tomatoes without pesticides and samples impregnated with chlorantraniliprole and flonicamid pesticide, were acquired by both the low cost and Pika L cameras. The Regions of Interest (ROIs) were determined and the averaged spectral value of these ROIs were calculated as the representative spectrum of each of them. Finally, the following classification models were performed with 720 ROIs: support vector classifier (SVC), K nearest neighbor, decision trees and multilayer perceptron (MLP). Furthermore, feature selection was carried out to select the main variables or wavelength bands. According to the values obtained for accuracy, recall, f1-score and precision, the best model for chlorantraniliprole detection was the MLP (accuracy=0,9). The preliminary results confirmed the feasibility and effectiveness of hyperspectral imaging to detect pesticide residues on tomatoes.