Simultaneous freshness and ambient humidity sensing in agri-fresh food using a fluorescent sensor array and DCNN

Water vapor severely interferes with colorimetric/fluorescent gas sensors, undermining their accuracy in real-world applications. A universal, customizable framework for humidity compensation remains elusive. To address this challenge, a paradigm shift is proposed in which humidity responses are actively exploited for signal compensation rather than being suppressed. This study developed a fluorescent sensor array comprising 30 sensing units, establishing a hybrid composite architecture. This architecture integrates pH-indicating dyes (hemopyrrole hydrochloride, puerarin, fisetin) sensitive to spoilage-related volatile acidic/alkaline gases, combined with a dedicated riboflavin-based humidity label exhibiting highly reversible humidity response capabilities. This hybrid composite system combines organic fluorescent dyes with a PTFE microporous membrane support layer and a laminated polyethylene cover film, forming a multi-level, functionally integrated sensing module. This sensor module was affixed to the top-space inner surface of packaging for longan, button mushrooms, and snap beans. Fluorescence images captured with a smartphone were processed by a multitask deep convolutional neural network, enabling simultaneous, nondestructive tri-level classification of both freshness (fresh/sub-fresh/spoiled) and storage humidity (low/optimal/high). The model achieved 97.67% accuracy in freshness classification and 95.77% in humidity grading. Crucially, incorporating humidity compensation improved freshness prediction from 92.50% to 97.67%, substantially reducing misclassification under high-humidity conditions. This work offers a novel, lightweight, and intelligent approach for concurrent monitoring of humidity and quality in fresh produce supply chains.