Machine learning-enhanced color recognition of test strips for rapid pesticide residue detection in fruits and vegetables

Abstract

Food safety, particularly the risks posed by pesticide residues, has become a critical public health concern. Existing detection methods are often slow, expensive, and require complex equipment, limiting their widespread use. This study introduces a rapid test strips system for pesticide residues, focusing on cholinesterase and organophosphate pesticides. The system combines a colorimetric reaction with machine vision to automate image analysis. Key image processing techniques, including noise reduction and threshold extraction, are used to analyze RGB values from the test strips. Multicolor feature indices are then derived to process the data. Additionally, an improved genetic programming-symbolic regression (GP-SR) model is developed to establish the relationship between these indices and pesticide residue levels. Experimental results show that the enhanced GP-SR model increases the R² value by up to 0.195 after normalization, improves the coefficient of determination by 2.5%, and reduces the RMSE by 16%. This approach offers a more efficient and accurate method for detecting pesticide residues in fruits and vegetables, contributing to improved food safety monitoring.