Real-time freshness monitoring of fruits and vegetables integrating 3D-printed alginate-based colorimetric sensors with deep convolutional neural networks

Abstract

3D printing, as an innovative smart label fabrication technology, offers exceptional design flexibility and the ability to integrate diverse materials. In this study, sodium alginate (SA) hydrogel, polyvinyl alcohol (PVA), and glycerol were used as the film-forming matrix (ink), incorporating 12 pH-sensitive indicators. This ink was employed in 3D printing to produce colorimetric sensor arrays with varied layer numbers and fill densities. Considering preparation time, material consumption, and responsiveness to different concentrations of ammonia and acidic solutions, the sensor array with a 100 % + 60 % fill density demonstrated an optimal balance between gas permeability and color change intensity. By integrating these sensor arrays with deep convolutional neural networks (DCNNs), precise classification of microbial growth stages was achieved by detecting acidity and alkalinity changes in volatile metabolite gases produced during microbial growth. The MobileNetV2 model attained a classification accuracy of 81.41 % for E. coli, while the ShuffleNet and Xception models achieved 98.74 % accuracy for B. cereus. Moreover, these colorimetric sensor arrays with varying pH sensitivities were applied to monitor freshness changes in fruits and vegetables during storage. Results showed significant color changes within the first 48 h for strawberries, raspberries, Malanto, and soybean sprouts, after which colors stabilized. In mulberries, a sharp pH drop during the later spoilage phase induced the most pronounced color change. On day 2 of storage at 25 °C, the E sensor label on mulberries exhibited a color difference of 118.85, 6.77 times higher than that of the control group. Additionally, the H sensor label's color differences increased from 87.92 to 202.88 over days 2, 4, and 6, indicating a gradual formation of a strongly acidic environment inside the packaging. The trained MobileNetV2 model achieved freshness classification accuracies of 97.95 %, 99.49 %, 95.90 %, 97.44 %, and 93.85 % for strawberries, mulberries, raspberries, Malanto, and soybean sprouts stored at 4 °C, respectively, in the test set. This study presents an innovative, real-time monitoring solution for fruit and vegetable preservation and quality control.