A digital twin approach for enzymatic browning and shelf-life prediction in postharvest bananas using deep learning

Banana (Musa spp.) is a climacteric fruit with high postharvest perishability driven by enzymatic browning catalyzed by polyphenol oxidase (PPO). This study aimed to develop a digital twin system integrating a convolutional neural network (CNN) for image-based quality classification with a kinetic model of enzymatic browning for shelf-life prediction under retail storage conditions. A dataset of 1203 banana images (1024 x 1024 px) at consumption maturity (Class A: “uniform color, appearance and firmness”; Class B: "yellow color with small imperfections, light brown spots and slightly less firmness compared to class A”; Defect: “significant discoloration with multiple large brown spots and unacceptable firmness for consumption”) was used to train an EfficientNet-B3 with transfer learning, featuring a dual output: quality classification and continuous browning score (B-score “0: without browning” – “1: maximum browning”). This score initialized a Michaelis–Menten kinetic model coupled with an Arrhenius equation (K_m: 3.0 mM; E_a: 95.3 kJ/mol) simulating PPO-mediated browning dynamics over time, assuming free contact between the enzyme and the substrate. Retail conditions of 15, 20, and 30 °C at 67% RH were assessed. The CNN achieved 95.0% accuracy (MAE: 0.063). The digital twin showed that Class A bananas maintained acceptable browning (B-score: 0.5) for 7.0, 4.7, and 1.7 days at 15, 20, and 30 °C, respectively. Class B showed critically reduced shelf-life (B-score: 0.5) of 2.0, 1.2, and 0.4 days at 15, 20, and 30 °C, respectively, with commercial acceptability lost (B-score > 0.8) in under 1 day at 30 °C. Defect class bananas exceeded a B-score of 0.5 from day zero at all temperatures, confirming they must be redirected to industrial processing before reaching retail display. This CNN–digital twin framework is a novel food 4.0 tool for individualized real time shelf-life prediction at the point of sale, supporting inventory rotation, waste reduction, and quality management in the fruit supply chain.