Zhang Shasha, He Haibin, Han Xiaoxue, Gu Huayu, Zhang Shuaibo, Tang Zhaorun, Ke Xianwen, Liu Juhua, Liu Xinghai
Hubei Engineering Technology Research Center of Spectrum and Imaging Instrument, School of Electronic Information, Wuhan University, Wuhan 430072, China; Wuhan Institute of Quantum Technology, Wuhan 430206, China.
National Engineering Research Center for Multimedia Software, Institute of Artificial Intelligence, and Hubei Key Laboratory of Multimedia and Network Communication Engineering, School of Computer Science, Wuhan University, Wuhan 430072, China.
Food Res Int. 2025 Oct;218:116833. doi: 10.1016/j.foodres.2025.116833. Epub 2025 Jun 13.
The perishability of fruits and vegetables (F&v) presents a significant challenge in maintaining food quality and safety. However, current methods for monitoring the freshness of fresh-cut F&v remains limited. This study introduces a novel deep learning-based colorimetric indicator system designed for the nondestructive monitoring of freshness in fresh-cut F&v packed in polylactic acid (PLA) bags. The system employed an ethylcellulose-based indicator (EMT), which showed a distinct color transition in response to carbon dioxide (CO) levels (0 %-30 %) during storage. In addition to its sensitivity, the EMT exhibited remarkable stability and reusability. Moreover, using fresh-cut green pepper as a model, the relationship of 'physiological state-freshness-indicator color' was constructed through the application of feature extraction algorithms (PCA and FLDA) in machine learning for the first time. The correlation was harnessed in conjunction with deep learning algorithms for image recognition and analysis. This approach mitigated or eliminated recognition errors arising from individual differences in human visual perception and variations in shooting conditions. The results indicated that the system could accurately, quickly, and nondestructively assess the freshness of fresh-cut green pepper, and the average accuracy of MobileNetV3-Small recognition could reach 96.09 % under k-fold cross-validation. The proposed strategy offered a highly accurate, real-time, and nondestructive method for monitoring produce freshness, with potential applications in food safety, health monitoring, and environmental protection.
水果和蔬菜（F&v）的易腐性给维持食品质量和安全带来了重大挑战。然而，目前用于监测鲜切F&v新鲜度的方法仍然有限。本研究介绍了一种基于深度学习的新型比色指示系统，该系统设计用于无损监测包装在聚乳酸（PLA）袋中的鲜切F&v的新鲜度。该系统采用了一种基于乙基纤维素的指示剂（EMT），在储存期间，该指示剂对二氧化碳（CO）水平（0%-30%）呈现出明显的颜色变化。除了灵敏度外，EMT还表现出显著的稳定性和可重复使用性。此外，以鲜切青椒为模型，首次通过在机器学习中应用特征提取算法（PCA和FLDA）构建了“生理状态-新鲜度-指示剂颜色”之间的关系。利用这种相关性并结合深度学习算法进行图像识别和分析。这种方法减轻或消除了由于人类视觉感知的个体差异和拍摄条件的变化而产生的识别误差。结果表明，该系统能够准确、快速且无损地评估鲜切青椒的新鲜度，在k折交叉验证下，MobileNetV3-Small识别的平均准确率可达96.09%。所提出的策略为监测农产品新鲜度提供了一种高精度、实时且无损的方法，在食品安全、健康监测和环境保护方面具有潜在应用。
