关键词: 生姜；电子鼻；普适性；反向传播神经网络；集成学习算法

Abstract: To address the dual challenges of difficulty in detecting the spoilage levels of ginger in actual production and insufficient generalization capacity of detection models across multiple varieties, this study proposed a method for detecting spoilage levels of multiple ginger varieties based on electronic nose (E-nose) combined with a back propagation neural network (BPNN) optimized by the Boosting ensemble learning algorithm. To achieve universal detection of spoilage levels across various varieties of ginger, four varieties including Anqiu tender ginger, Shandong large ginger, Sichuan tender ginger and Yunnan small yellow ginger were selected for this study. The characteristic volatile gases produced at different spoilage stages of each variety were collected. Based on the collected sensor data, six features including baseline value, response amplitude, maximum value of the first-order derivative, transient value at the 10th second, variance, and rise time were extracted to establish the feature space. Random forest (RF), gradient boosting decision tree (GBDT), and BPNN were used to establish models for detecting ginger spoilage levels. The results showed that the BPNN model achieved the highest spoilage detection accuracy of 96.70% for Sichuan tender ginger. Given the potential for further performance improvement, the BPNN was optimized using Boosting and Bagging algorithms. The resulting Boosting-optimized BPNN model demonstrated superior performance, achieving a detection accuracy as high as 98.80% for Yunnan small yellow ginger. Furthermore, the model was validated on the remaining ginger varieties. The results showed that the accuracy for all varieties exceeded 90%. This study demonstrates that the Boosting-optimized BPNN model enables cost-effective and highly efficient detection of the spoilage of multiple ginger varieties, which holds practical significance for subsequent research on E-nose and ginger.

Key words: ginger; electronic nose; universality; back propagation neural network; ensemble learning algorithm