Development of pesticide residues detection based on Hyperspectral image using mobile phone cameras

Abstract

Pesticides have been the most frequently used chemicals in recent decades to protect agricultural commodities from pests. However, pesticide overuse has impacted the environment and human health. Thus, early pesticide residue detection can effectively reduce pesticides' negative effects. This study proposed a rapid detection method for cypermethrin (CYP) residues using a low-cost hyperspectral based on a mobile phone device. Furthermore, this research was conducted on filter papers with 5 concentrations of CYP (i.e., 0, 175, 500, 1000, 10,000 mg/L). Then, the measurement was done under a lighting box with 3 light sources, i.e., visible, UV, and NIR, that emitted specific wavelengths of 400-700, 380, and 850 nm, respectively. In this study, digital filters (400-700 nm) were employed to approach the acquisition mode of hyperspectral, known as “Area-scanning.” Subsequently, data processing steps were performed to extract CYP residue information from the filter papers’ photos. In the model establishment process, all data were randomly divided into a training set of 80% and a testing set of 20%. Furthermore, the low-cost hyperspectral technique was coupled with principal component analysis (PCA) and three classifier models, which were linear discriminant analysis (LDA), support vector machine (SVM), and artificial neural network (ANN). Three different smartphone resolutions (i.e. 13, 32, and 64 MP) impacted the reflectance result of digital filters, with the 32 MP resolution giving the most apparent details. Finally, significant classification accuracies of cypermethrin residues reached 75%, 81.2%, and 84.4% of LDA, SVM, and ANN, respectively. The addressed method has great potential to detect pesticide residues on-site as a non-destructive, cheap, and easy-to-use method.