TTI and pH-responsive dual colorimetric sensor arrays combined with a cascaded deep learning approach for dynamic monitoring of freshness of fresh-cut fruits

Huijie Jia; Chenlin Wu; Min Huang; Qibing Zhu; Min Zhang

doi:10.1016/j.foodchem.2025.145495

TTI and pH-responsive dual colorimetric sensor arrays combined with a cascaded deep learning approach for dynamic monitoring of freshness of fresh-cut fruits

Food Chem. 2025 Jul 8;492(Pt 2):145495. doi: 10.1016/j.foodchem.2025.145495. Online ahead of print.

Authors

Huijie Jia¹, Chenlin Wu¹, Min Huang², Qibing Zhu¹, Min Zhang³

Affiliations

¹ Key Laboratory of Advanced Process Control for Light Industry (Ministry of Education), Jiangnan University, 214122 Wuxi, China.
² Key Laboratory of Advanced Process Control for Light Industry (Ministry of Education), Jiangnan University, 214122 Wuxi, China. Electronic address: huangmin2004@jiangnan.edu.cn.
³ State Key Laboratory of Food Science and Resources, Jiangnan University, 214122 Wuxi, Jiangsu, China; China General Chamber of Commerce Key Laboratory on Fresh Food Processing & Preservation, Jiangnan University, 214122 Wuxi, Jiangsu, China.

PMID: 40651146
DOI: 10.1016/j.foodchem.2025.145495

Abstract

Dynamic shelf-life monitoring of fresh-cut fruits faces challenges from temperature fluctuations and packaging failures in cold chains, causing discrepancies between theoretical predictions and actual spoilage. This study developed a dual colorimetric sensor array combining pH-responsive indicators and time-temperature integrators (TTIs) to overcome single-parameter limitations. The methylcellulose-based system integrates (1) CO₂-sensitive pH dyes (methyl red/bromocresol green) tracking freshness via pH changes, and (2) TTI elements using temperature-dependent HAuCl₄-l-ascorbic acid reactions that generate gold nanoparticle color transitions. Packaging atmosphere shifts during storage-such as CO₂ buildup-activate the sensors. The array exhibits strong structural and environmental stability. A cascaded deep learning framework combining object detection (YOLOv8), feature extraction (ResNet-18), and Bayesian modeling achieves 93.3 % accuracy in predicting mango/kiwifruit shelf life-35 % and 18 % improvements over standalone TTI and pH sensors, respectively. The system supports real-time smartphone-based monitoring with automated alerts, offering a practical tool for cold chain freshness tracking.

Keywords: Deep learning; Kinetic model; Methylcellulose; Shelf-life dynamic monitoring; TTI; pH indicators.