Improving the detection accuracy of the nitrogen content of fresh tea leaves by combining FT-NIR with moisture removal method

The nitrogen content (NC) is one of the critical indicators of tea quality, and many studies have been conducted using NIR spectroscopy to determine tea constituents. However, this method has been found to have limited accuracy for component estimation because the spectra are affected by moisture in the samples. In this study, external parameter orthogonalization (EPO) was introduced to filter out the effect of moisture in fresh tea leaves on NIR spectra. Then, a feature selection algorithm was applied to determine the optimal NC wavelength to improve the prediction precision. Finally, a partial least squares (PLS) prediction model was established. The PLS model based on EPO and VCPA-IRIV achieved satisfactory prediction results, with an increase in Rp2 to 0.9371 from 0.5846 for the full spectral PLS model without treatment. Overall, this study found that eliminating the effect of moisture on spectra could improve detection accuracy of the model significantly.