Classification of oolong tea varieties based on hyperspectral imaging technology and BOSS‐LightGBM model

Abstract

AbstractA fast and nondestructive detection method based on hyperspectral imaging technology (HSI) was investigated in this study to discriminate different oolong tea varieties. Five varieties of oolong tea were taken as the research object. Multiplicative scatter correction was used to reduce the influence of noise in the raw spectra. Then competitive adaptive reweighted sampling and bootstrapping soft shrinkage (BOSS) were applied, respectively, to select characteristic wavelengths. Extreme gradient boosting (XGBoost) and light gradient boosting machine (LightGBM) were individually utilized to establish classification models. Finally, the BOSS‐LightGBM model for discriminating tea varieties achieved the best performance, with the accuracy of 100% in the training set and 97.33% in the prediction set. Therefore, it is feasible to use HSI technology coupled with the BOSS‐LightGBM model for the classification of oolong tea varieties.Practical applicationsTieguanyin tea is a high value commodity in the tea market. Replacing Tieguanyin tea with cheaper oolong tea varieties is a common way utilized by illegal traders to maximize profit. Traditional methods for identifying tea varieties are time‐consuming and destructive, and are thus unable to meet the requirements of modern agriculture. In this study, hyperspectral imaging technology (HSI) was applied to realize the fast and nondestructive detection of tea varieties. The final results show that using HSI technology to discriminate different oolong tea varieties is feasible, and also provide a theoretical basis for the design of a portable tea variety detection device.