Pesticide residues on agricultural produce pose considerable health risks, necessitating the development of rapid, accurate, and accessible detection solutions to enhance food safety. This paper presents the design and implementation of a portable, low-cost system for onsite pesticide residue detection in bananas. The proposed system is built around a Raspberry Pi microprocessor integrated with a Pi camera for image acquisition and a chemical sensor for volatile organic compound (VOC) analysis. A hybrid deep learning framework comprising Convolutional Neural Networks (CNNs) and Support Vector Machines (SVMs) is employed, trained on a custom dataset consisting of 4,800 images and 4,800 VOC samples obtained from both untreated and pesticide-exposed bananas under natural lighting conditions. The embedded system performs real-time data acquisition and processing, achieving detection accuracies of 94.20% using CNN and 96.00% using SVM, with an average inference time of approximately 1.60 seconds per sample. Detection results are displayed on a 16 × 2 LCD module, and an alert is issued via a buzzer when pesticide concentrations exceed a predefined safety threshold. Experimental validation confirms the system’s effectiveness in combining visual and chemical data for improved detection accuracy. Future work will focus on enhancing sensor specificity and optimizing the classification models to support broader agricultural applications. Ground truth validation of pesticide residues was performed using an Agilent 7890A GC system with 5975C Mass Selective Detector, confirming residue thresholds based on WHO Maximum Residue Limits.
Devarajan et al. (Sat,) studied this question.