An effective feature reduction method is a key issue to improve the detection performance of the electronic nose (e-nose). In this study, a feature reduction method coupled with a support vector machine (SVM) was proposed to enhance the detection performance of the e-nose for the quality detection of tea. Firstly, the time-domain features were extracted, which can represent the original gas information of different grades of tea. Secondly, to consider the importance of the relationship between each feature and output category, a subset of multiple features with the best variable importance of projection (VIP) score was generated to obtain the optimized feature set. Finally, kernel principal component analysis (KPCA) and kernel entropy component analysis (KECA) were performed to further reduce the correlation between features to obtain the best feature set. The results indicated that VIP-KECA can obtain the best feature set effectively, and a good classification accuracy of 98% was obtained. This study shows that the feature reduction method is effective for enhancing the detection performance of the e-nose. It also provides an effective technique to monitor the quality of tea.

No abstract