Visualizing distribution of moisture content in tea leaves using optimization algorithms and NIR hyperspectral imaging

Abstract Effective visualization of moisture content in tea leaves is very important in the tea cultivation industry and in favor of irrigation management in tea garden. In order to obtain the moisture content distribution map of tea leaves, successive projections algorithm (SPA) coupled with stepwise regression (SPA-SR) and competitive adaptive reweighted sampling (CARS) coupled with stepwise regression (CARS-SR) were proposed to select characteristic wavelengths in this study. The whole region of the tea leaves were selected as region of interest (ROI) to extract the NIR hyperspectral reflectance. Moreover, Savitzky-Golay smoothing (SG), orthogonal signal correction (OSC), multiplicative scatter correction (MSC) and detrending were used to handle with raw spectra. In addition, four feature selection algorithms (SPA, CARS, SPA-SR and CARS-SR) were used to extract the most effective wavelengths. Furthermore, multiple linear regression (MLR) was adopted to establish the prediction models based on spectrum after 20 different combination algorithm treatments. The results showed that SPA-SR and CARS-SR can effectively improve the correlation coefficient of prediction set in established MLR models compared with SPA and CARS, respectively. Besides, the combination algorithm for obtaining the best prediction MLR model was SG-MSC coupled with CARS-SR ( R p 2  = 0.8631 and RMSEP = 0.0163), and it was applied to retrieve the distribution of moisture content in tea leaves. Visualizing distribution map of tea leaves offered a more intuitive and comprehensive assessment of moisture contents at each pixel, and it provides a novel approach to evaluate plant irrigation.