This article (open access) reports the development of a classification model for the country of origin of black tea. The authors collected both FTIR and NIR spectral data from a reference library of 360 black tea samples sourced from prominent tea cultivation regions across the world, including China, Darjeeling (India), Assam (India), Sri Lanka, Kenya, Ethiopia, Burundi, and Malawi. They compared different machine learning models to build a predictive classification system. They found that the best results were obtained when SNV and 1DER spectral pre-processing methods were also used. They concluded that all of the machine learning models gave superior prediction performance compared to traditional PLS-DA modelling, with the best giving a classification accuracy of 100 %. They also identified and validated a set of significant wavenumber regions in FTIR and NIR spectra for discriminating black tea GI regions. The authors conclude that the developed workflow is a novel, rapid, easy to operate, cost-efficient, and non-destructive method, and can be regarded as a “green analytical technique” since no solvents and reagents are used during the process. It has the potential to form the basis of an on-site, real-time solution for Geographic Origin inspections throughout the entire supply chain, particularly within developing countries where tea cultivation is prominent. However, further work is required to enhance the size and diversity of the database and validate the model’s transferability between devices.
Photo by charlesdeluvio on Unsplash
Comments