Deep Learning-Based Prediction and Regulation for Remaining Shelf Life of Blueberries

The rapid quality deterioration of blueberries during storage underscores the need for precise prediction and regulation of remaining shelf life. This study proposes a novel closed-loop framework integrating an analytic hierarchy process-principal component analysis-long short-term memory (AHP-PCA-LSTM) architecture to estimate remaining shelf life, coupled with Gaussian process regression (GPR) for temperature-based regulation. Utilizing a primary dataset of 72 observations (6 temperatures × 12 sampling points) expanded to 600 via third-order spline interpolation to capture nonlinear kinetics, blueberries were individually packaged in breathable biodegradable films and stored across diverse temperatures (4-25°C) at 50% ± 2% relative humidity. The model employs an autoencoder to construct a "freshness" index (0-100), where a failure threshold of 30 was empirically validated against a 10% fruit decay rate and sensory rejection points. The results indicate that the AHP-PCA-LSTM model effectively reduces data redundancy, with catalase identified as the primary quality weight (0.34), and achieves high-precision remaining shelf life forecasting with a 2.5% error rate, significantly outperforming traditional Arrhenius kinetic models. Furthermore, GPR with Bayesian optimization successfully mapped the nonlinear freshness-temperature dependency, enabling the inverse deduction of optimal storage conditions to prolong freshness. This dynamic system offers a robust paradigm for advanced blueberry supply chain management by quantifying predictive uncertainty through 95% confidence intervals. PRACTICAL APPLICATIONS: This study provides a tool for blueberry producers and sellers to predict how long fruit will stay fresh. By adjusting storage temperatures based on our model, companies can reduce food waste and ensure better quality for consumers.