Machine learning-assisted ZnO-based sensor for multi-species recognition of volatile aroma components in tea plant

Abstract

The aroma profile of plant (e.g. tea plant) is mainly influenced by various aromatic substances such as leaf alcohol and geraniol, which play a crucial role in determining the quality of growth and can be also served as biomarker to evaluate the infestation of pests and diseases. Therefore, the detection of volatile aroma components is of great significance to assess the quality and monitor the pests and diseases in plant. In this work, ZnO-based sensor is fabricated to investigate its gas-sensing performance towards six types of representative tea aromas (leaf alcohol, geraniol, capraldehyde, octanol, phenethyl alcohol, methyl salicylate). As a result, the ZnO-based sensor shows the highest gas-sensing response (∼110) with a fast response/recovery time of 29 s / 7 s for 10 ppm leaf alcohol at 325 ℃, and exhibits an impressive limit of detection for leaf alcohol as low as 0.5 ppm with a gas-sensing response value of 6. Meanwhile, machine learning algorithms (SVM, WNN, KNN, LDA, CART and NB) are applied to achieve the accurate recognition of the types and concentrations for tea aromas based on the gas-sensing response values of six types of tea aromas at 225 ℃, 275 ℃ and 325 ℃. The highest classification accuracy can reach 95.8 % and the predication accuracy for the concentration of leaf alcohol is about 97.8 %. This work assists the combination of machine learning with gas sensor in the detection and recognition of multi-species gases, providing supports for the early diagnosis and warning of pests and diseases in plant.