Blueberry Remaining Shelf-Life Prediction Based on the PSO-CNN-BiLSTM-MHA Model

Remaining shelf-life prediction of blueberries is crucial in reducing economic losses and enhancing market competitiveness. To meet the demand for high accuracy in predicting the remaining shelf-life of blueberries, this paper proposes a fusion model (PSO-CNN-BiLSTM-MHA) that combines Particle Swarm Optimization (PSO), Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory network (BiLSTM), and Multi-Head Attention (MHA) mechanisms for predicting the remaining shelf-life of ‘Emerald’ blueberries under two temperature conditions,4∘ Cand25∘ C. In this study, seven key features were selected from fifteen quality indicators of blueberries using the embedded method, and the PSO algorithm was used to determine the optimal hyperparameter combination of the model, which effectively improved its prediction performance. The experimental results show that our model outperforms other models in all evaluation metrics under both temperature conditions and demonstrates excellent prediction ability and stability. This study provides an effective technical approach for the accurate prediction of the remaining shelf-life of blueberries and other fruits.