Application of Raman Spectroscopy Coupled With Chemometrics for the Detection and Quantification of Mancozeb Residues in Collard Green

Abstract

The presence of pesticide residues in food crops poses serious health concerns, necessitating precise, rapid and accessible detection techniques. This study investigates the use of Raman spectroscopy combined with advanced data analysis techniques to detect and quantify Mancozeb residues in collard greens. The primary objective was to evaluate the viability of this approach for accurate pesticide residue monitoring in leafy vegetables. Raman spectral data were collected and preprocessed using a standard normalization technique to reduce spectral noise and enhance signal quality. Dimensionality reduction was achieved through a statistical method that extracted key spectral features and successfully differentiated control from treated samples, explaining a combined variance of 86% across the first two principal components. Graphical score plots revealed clear clustering patterns across various residue concentrations, ranging from 0.01 to 0.5 parts per million, with samples categorized according to regulatory residue limits. To further assess predictive capability, several machine learning models were developed for classification and quantification, including deep learning–based and ensemble-based approaches. Among these, the support vector model achieved the highest classification precision of 95% and demonstrated strong calibration and prediction accuracy. A convolutional neural network achieved 99% training accuracy and 98% testing accuracy, effectively recognizing complex spectral patterns. Statistical validation using analysis of variance confirmed that the observed model differences were significant, supporting the robustness of the selected algorithms. The proposed method accurately quantified Mancozeb residues within the tested range and demonstrated high sensitivity even at low concentration levels. This study highlights the potential of Raman spectroscopy, integrated with computational modelling, as a non-destructive, fast and cost-effective tool for pesticide residue detection in food safety applications.