AI-driven design of thymol-based composite films for cucumber preservation

To address uncontrollable active compound release, high trial-and-error costs, and low design efficiency in traditional active food packaging research, deep learning was introduced into active packaging design in this study, focusing on precise design of thymol-based composite films and their application in cucumber preservation. First, a dataset with 99 samples was constructed, covering 9 input variables and 1 output variable. A 4-layer fully connected neural network with linear/nonlinear activation functions was used to train a prediction model for the "minimum effective concentration" of active compounds. The model’s R² reached 0.84, explaining 84 % of the dependent variable variation. Under this study’s film-forming conditions, the model predicted thymol’s minimum effective addition amount as 0.1245 % (w/w). Based on this, composite films were prepared with actual thymol additions of 0.2 %-0.4 % (w/w). Results showed these films exhibited excellent mechanical, thermal, barrier, antioxidant, and antibacterial properties, with the 0.4 % thymol film reducing Escherichia coli and Staphylococcus aureus counts by 3 and 2.5 log values, respectively. When applied to cucumber preservation at 4 °C for 15 days, they reduced weight loss by 35 % and total color difference by 26 % compared to unpackaged controls, effectively delayed hardness decline, regulated oxidase (POD, SOD, CAT) activities, and significantly improved preservation effect. In conclusion, this study confirmed deep learning could realize predictive modeling and optimization of active packaging components, providing a standardized and efficient R&D framework for active food packaging with broad industrial application potential.