Chiang Mai Journal of Science

     The demand for fresh fruit juices is increasing, however, some food processors are using spoiled fruit for juice production in order to seek higher profits. Electrical impedance spectroscopy (EIS), in combination with machine learning, enables wide applicability in food quality inspection, as it does not necessitate expensive instruments or complex sample processing. However, EIS data vary with temperature changes. The measurement of EIS data requires consistent temperature conditions. To overcome this limitation, data augmentation was applied to train the model based on EIS to improve the performance and robustness of the model. The training set consists of 200 EIS data measured under consistent temperature conditions and 200 EIS data with added noise and a recognition model for detecting spoiled apple juice was established. Under inconsistent temperature conditions, the model’s accuracy in identifying spoiled fruit juice reached 98% on the test set, while the accuracy dropped to 50% without data augmentation. This study demonstrates that the application of data augmentation on the training set reduces the need for consistent temperature conditions during the collection of EIS data, thereby improving the model’s robustness and eliminating the waiting time for data stability. Therefore, applying data augmentation to EIS and machine learning provides a rapid, practical, and reliable method for assessing the quality of liquid products.