Image‐Based Modeling and Prediction of Banana Decay Under Blue Light Irradiation

ABSTRACT To address the issue of postharvest spoilage and deterioration in bananas, this study proposes an innovative method that integrates blue laser preservation with dynamic adaptive image prediction. This approach overcomes the limitations of traditional preservation methods, such as nutrient loss and chemical residues, by developing a green and efficient preservation technology. A 470 nm blue laser was applied to quantitatively analyze the weight loss rate, firmness variation, and browning degree of bananas. A three‐dimensional convolutional neural network (3D‐CNN) incorporating a squeeze‐and‐excitation (SE) module was constructed to localize decay regions through spatiotemporal feature extraction accurately. Furthermore, a dynamically adaptive weighted loss function was designed, which adjusts weights based on image gradients and second‐order derivatives to enhance prediction consistency. Under 60 PPFD blue light irradiation, the banana weight loss was reduced by 32.5%, and softening was delayed by 48 h. On the fourth day, the decay prediction model achieved a PSNR of 72.66 (an improvement of 11.01% compared to ConvLSTM) and an MAE reduced to 0.026 (a 55.1% decrease in error). This method achieves closed‐loop optimization of preservation treatment and decay prediction, providing a novel paradigm for intelligent preservation of perishable foods.