Abstract: To investigate the quality changes and shelf-life of blueberries stored under different temperature conditions, ‘Yikemei’ blueberries were evaluated for several quality indicators including soluble solid content, mass loss rate, decay incidence, and texture parameters during storage at 5, 10, 15, 20, and 25 ℃. Feature selection was performed using a binary gray wolf optimization algorithm, identifying seven key features influencing the shelf life as input variables for modeling. A shelf-life prediction model of blueberries was developed by the combined use of the pied kingfisher optimizer (PKO), convolutional neural network (CNN), bidirectional long short-term memory (BiLSTM) and attention mechanism (AT). In the CNN-BiLSTM-AT network, parameter optimization was conducted using the PKO to determine the optimal learning rate, regularization parameters, attention key values, and the number of BiLSTM neurons. The results indicated that compared with the CNN-LSTM model, the PKO-CNN-BiLSTM-AT model exhibited 76.13%, 80.96%, 92.03%, and 71.75% reductions in mean absolute error, mean absolute percentage error, mean squared error, and root mean squared error, respectively, while the coefficient of determination (R²) increased by 5.85%. These findings demonstrated that the introduction of PKO significantly improved the predictive performance of the CNN-BiLSTM-AT model. This study provides theoretical support for the shelf-life prediction of blueberries stored under different temperature conditions.

Key words: blueberry quality indicators, binary grey wolf optimization, pied kingfisher optimizer, deep learning model, shelf-life prediction