Estimating leaf chlorophyll contents of shade grown tea using hyperspectral data

Appearance, aroma and taste are important factors for assessing the quality of tea (Camellia sinensis) and then shading of tea is performed to increase chlorophyll content, which is an important factor for evaluating the appearance and good taste. Although some traditional approaches that require tremendous efforts for the collection of samples and laboratory chemical analyses have been applied, they are not feasible for long-term monitoring. In contrast, hyperspectral remote sensing is proven to be an efficient way for chlorophyll content monitoring. In this study, the three different approaches of kernel-based extreme learning machine (KELM), random forests (RF), and deep belief nets (DBN) were compared to assess the potential for estimating leaf chlorophyll contents from hyperspectral data with existing supervised learning models. Overall, regression models based on KELM yielded the highest performance, achieving a Root Mean Square (RMS) error of 0.20-0.56μg/cm2.