Assessing tea foliar quality by coupling image segmentation and spectral information of multispectral imagery

Abstract

In-situ rapid detection of biophysical parameters in tea leaves using spectral data is essential for enhancing the quality and yield of tea. However, a major challenge with the current application of spectral technology is its inability to completely distinguish between old leaves and picked leaves within the field of view, which affects the accurate correspondence of biochemical elements. Therefore, this study achieved precise matching of biophysical parameters with spectral information by focusing on the spectra of picked leaves. By combining the Excess Green minus Excess Red (ExGR) with the image segmentation methods of Otsu and P75, the spectral features of picked leaves were effectively identified from complex backgrounds. Additionally, the vegetation indices (VIs) closely associated with the biophysical parameters of tea were selected, and a partial least squares regression (PLSR) model was applied for parameter inversion. Results demonstrated that the VIs calculated using Otsu (VI_OtsuPix) and P75 (VI_P75Pix) exhibited significantly improved correlations with the biophysical parameters of tea compared with those calculated using ExGR > 0 (GreenPix). The PLSR model based on VI_OtsuPix performed well in estimating the total polyphenols (TPP), achieving a coefficient of determination (R2) of 0.39 and a root mean square error (RMSE) of 32.24 mg g−1. In predicting free amino acids (FAA), VI_P75Pix demonstrated the best inversion accuracy (R2 = 0.53, RMSE = 3.41 mg g−1). These findings not only confirmed the potential of integrated image technology in the non-destructive assessment of biophysical components in picked leaves but also provide the tea production and processing industry with a fast and cost-effective method for quality monitoring.