An efficient tea quality classification algorithm based on near infrared spectroscopy and random Forest

Abstract

Traditional tea quality evaluation methods are based on chemical testing, such as gas chromatography‐mass spectrometry (GCMS) and high‐performance liquid chromatography (HPLC). However, the process of extracting chemical components is generally time‐consuming and expensive, which makes it unsuitable for wide range of applications. Therefore, this paper presents a new approach to evaluate tea quality based on Near‐infrared Spectroscopy (NIRS) devices. In our method, factor analysis compression algorithm is first applied to initially compress the input NIRS vectors, which are acquired from tea samples with high dimensional data. Then, random forest algorithm is employed to construct a voting strategy. More precisely speaking, we proposed a low‐cost and convenient tea quality estimation scheme that can be widely used in tea industry. The proposed approach has been verified using tea NIRS datasets which were acquired from Fujian Province. Experiments show that the proposed NIRS‐based approach significantly outperforms the GCMS‐based and HPLC‐based methods. Specially, we achieved a highly competitive performance (AP = 0.989) on the comprehensive data set that contains 869 annotated Chinese tea samples, which means that tea quality can be estimated in a convenient and cheaper way.Practical ApplicationsThe proposed tea classification approach based on artificial intelligence which lend new perspectives to tea merchants and consumers insight and decision‐making. The approach can perform preference adjustments in various conditions such as regions, crowd habits, seasons, etc.

Fast Detection of Diarrhetic Shellfish Poisoning Toxins in Mussels Using NIR Spectroscopy and Improved Twin Support Vector Machines

“…To compare the performance of the proposed ITWSVM model with other state-of-the-art models in DSP toxins detection, we have conducted experiments under various IR. Multi-layer perceptron (MLP) (Murlidhar et al, 2021), extreme gradient boosting (XGBoost) (Zhao et al, 2022), adaptive boosting (AdaBoost) (Liu, 2010), random forest (RF) (Chen et al, 2021), and k-nearest neighbor (KNN) (Wang et al, 2021) models have been selected for comparison.…”

Classification of Tea Leaves Based on Fluorescence Imaging and Convolutional Neural Networks

“…Meanwhile, this process is time-consuming and expensive, which prevent its use by customers. Methods e.g., Vis-NIR spectroscopy [ 7 , 8 , 9 , 10 ], Fourier transformed infrared spectroscopy [ 11 , 12 ], laser-induced breakdown spectroscopy [ 13 ], chemical analysis [ 14 ], X-ray fluorescence spectroscopy [ 15 ], electronic nose [ 16 ], and liquid chromatography [ 17 ] have all been applied to tea classification. Although the classification abilities of these methods are good, drawbacks of high costs and large volumes of the experimental systems limit their utilization to laboratory experiments.…”

Green tea grades identification via Fourier transform near‐infrared spectroscopy and weighted global fuzzy uncorrelated discriminant transform

“…For example, Liu et al (2020) used competitive adaptive reweighting algorithms‐stepwise regression analysis combined with NIR spectroscopy to classify five kinds of Chinese tea and achieved satisfactory results. Hazarika et al (2018) developed a set of tea NIR detection devices, which realizes the functions of fixing, drying, grinding, weighing, and detection; Chen et al (2021) established a tea quality evaluation system with superior performance using NIR spectroscopy and random forest algorithm and realized the accurate classification of tea quality. NIR spectroscopy combined with the support vector data description algorithm successfully identified black tea from four countries (Ning et al, 2016).…”