RGB Imaging Based Estimation of Leaf Chlorophyll Content

Abstract

Leaf chlorophyll content (LCC) is an important indicator of plant health. It can reveal for instance whether the plant has received too much or too little nutrients (nitrogen in particular). Corrective measures based on early diagnosis of nitrogen deficiency can prevent yield reduction. Optical approaches to estimating LCC are typically based on measuring the reflectance of leaves (close-range sensing) or canopy (remote sensing) with a spectrophotometer or a hyperspectral sensor. Given the cost of such devices, the use of colour cameras for LCC estimation has recently gained interest. However, existing approaches are mostly designed for very close range sensing, i.e. they allow the measurement of LCC for a single plant at a time, which limits their usefulness in practice. Furthermore, they do not exploit the anisotropic reflective properties of leaves and canopies. To investigate the feasibility of RGB imaging for remote sensing-based estimation of canopy LCC, we present a simulation based on a plant canopy reflectance model. An RGB camera model was used to generate RGB images with various reference illuminants and from 10 different viewing angles. We then applied linear and neural network based regression to predict LCC from RGB values without white balance. Results indicate a significant potential to use RGB sensors for remote sensing LCC estimation, particularly with a multi-angle approach.