Generalisation of tea moisture content models based on VNIR spectra subjected to fractional differential treatment

Abstract

Model generalisation for the detection of tea moisture content was investigated across different leaf surface orientations and tea varieties in this study. The micromorphology of leaves plucked from three tea bushes was analysed, and differences between different surface orientations and varieties were observed. The VNIR spectra (350–2500 nm) of the leaves were collected and analysed. Excellent prediction performance was obtained for moisture detection models based on spectra for the same leaf surface orientation and variety. By contrast, the prediction performance decreased severely if the test spectra were obtained for different leaf surface orientations and varieties. To solve this issue, differential treatments with fractional order between 0 and 2 were carried out on the spectra. The results showed that the prediction performance improved for generalisation between varieties and orientations, especially for orders of 0.4 or 0.6. The mechanism by which the fractional differential treatment mines the common information from the spectra with varying characteristics was elucidated by calculating the correlation coefficients between the moisture content and the spectra treated with different differential orders. The results of this study advance tea moisture detection based on VNIR spectra and the ability to generalise across spectra with different characteristics.