A Novel Kohonen One-Class Method for Quality Control of Tea Coupled with the Artificial Lipid Membrane Taste Sensors
The effective outlier detection method is an important technical in the process of product quality monitoring. In this work, a novel Kohonen one-class model coupled with the artificial lipid membrane taste sensors was proposed for tea quality control. Firstly, the taste information of five different grades of Tieguanyin tea was obtained based on the artificial lipid membrane sensors. Secondly, an improved Kohonen model based on the average Hausdorff distance was proposed to enhance the matching degree and transform multi-class model into one-class model. Thirdly, a reject ratio was proposed to optimize the domain boundary, which affected the classification performance. Finally, F1-score was introduced to evaluate the classification performance of one-class model. The results indicated that the F1-score for five different grades of Tieguanyin tea were 0.952, 0.947, 0.952, 1 and 0.947 respectively, and the average F1-score of 0.9596 was obtained. This study shows that the novel Kohonen one-class model coupled with the artificial lipid membrane taste sensors can be used as an effective outlier detection method to control the tea quality.