Intelligent vegetable freshness monitoring system developed by integrating eco-friendly fluorescent sensor arrays with deep convolutional neural networks

Abstract

Given the perishable and seasonal nature of vegetables, monitoring their freshness is essential to ensure food safety and reduce waste. Currently, there are limited packaging systems for fresh vegetables that incorporate intelligent freshness monitoring labels. Herein, we report on the development and application of a 3 × 6 fluorescent sensor array that exhibits pH-sensitive properties, utilizing curcumin, puerarin, and fisetin. During spoilage, yardlong beans and spinach, which had high protein content, produced alkaline volatile organic compounds (VOCs), whereas sweet corn, rich in sugar, emitted acidic VOCs. The fluorescent sensor array, integrated with deep convolutional neural network (DCNN), enabled non-destructive, real-time, and accurate classification of the freshness of the aforementioned three vegetables by detecting the acidity or alkalinity of their VOCs. The trained ResNet50 DCNN model achieved an overall accuracy of 96.21 % in classifying the freshness of the aforementioned vegetables in the testing set, with specific accuracies of 98.58 % for yardlong beans, 97.15 % for spinach, and 92.89 % for sweet corn, respectively. This intelligent freshness monitoring platform is adaptable for monitoring and classifying the freshness of a wide range of agricultural and food products.