Han Zhang, Ahmad Waqas, Rong Yanna, Chen Xuanyu, Zhao Songguang, Yu Jinghao, Zheng Pengfei, Huang Chunchi, Li Huanhuan
School of Mechanical Engineering, Jiangsu University, Zhenjiang 212013, China.
School of Food and Biological Engineering, Jiangsu University, Zhenjiang 212013, China.
Foods. 2024 May 30;13(11):1721. doi: 10.3390/foods13111721.
The oxidation step in Oolong tea processing significantly influences its final flavor and aroma. In this study, a gas sensors detection system based on 13 metal oxide semiconductors with strong stability and sensitivity to the aroma during the Oolong tea oxidation production is proposed. The gas sensors detection system consists of a gas path, a signal acquisition module, and a signal processing module. The characteristic response signals of the sensor exhibit rapid release of volatile organic compounds (VOCs) such as aldehydes, alcohols, and olefins during oxidative production. Furthermore, principal component analysis (PCA) is used to extract the features of the collected signals. Then, three classical recognition models and two convolutional neural network (CNN) deep learning models were established, including linear discriminant analysis (LDA), k-nearest neighbors (KNN), back-propagation neural network (BP-ANN), LeNet5, and AlexNet. The results indicate that the BP-ANN model achieved optimal recognition performance with a 3-4-1 topology at pc = 3 with accuracy rates for the calibration and prediction of 94.16% and 94.11%, respectively. Therefore, the proposed gas sensors detection system can effectively differentiate between the distinct stages of the Oolong tea oxidation process. This work can improve the stability of Oolong tea products and facilitate the automation of the oxidation process. The detection system is capable of long-term online real-time monitoring of the processing process.
乌龙茶加工过程中的氧化步骤对其最终的风味和香气有显著影响。本研究提出了一种基于13种金属氧化物半导体的气体传感器检测系统，该系统对乌龙茶氧化生产过程中的香气具有很强的稳定性和敏感性。气体传感器检测系统由气路、信号采集模块和信号处理模块组成。传感器的特征响应信号显示，在氧化生产过程中会快速释放出挥发性有机化合物（VOCs），如醛类、醇类和烯烃类。此外，主成分分析（PCA）用于提取采集信号的特征。然后，建立了三种经典识别模型和两种卷积神经网络（CNN）深度学习模型，包括线性判别分析（LDA）、k近邻（KNN）、反向传播神经网络（BP-ANN）、LeNet5和AlexNet。结果表明，BP-ANN模型在pc = 3时采用3-4-1拓扑结构实现了最佳识别性能，校准和预测的准确率分别为94.16%和94.11%。因此，所提出的气体传感器检测系统能够有效区分乌龙茶氧化过程的不同阶段。这项工作可以提高乌龙茶产品的稳定性，并促进氧化过程的自动化。该检测系统能够对加工过程进行长期在线实时监测。
