Estimation of ground canopy cover in agricultural crops using downward-looking photography

Fast and accurate estimates of canopy cover are central for a wide range of agricultural applications and studies. Visual assessment is a traditionally employed method to estimate canopy cover in the field, but it is limited by the costs, subjectivity and non-reproducibility of the produced estimates. Digital photography is a low-cost alternative method. In this study we tested two automated image classification methods, the first one based on a histogram-analysis method (Rosin), the second one based on a combination of a visible vegetation index and the L*a*b* colour space conversion (LAB2), which have both been previously tested in forestry, and a supervised image classification method (Winscanopy), to estimate canopy cover from downward-looking images of agricultural crops. These methods were tested using artificial images with known cover; this allowed exploring the influence of canopy density and object size on canopy cover estimation from photography. The Rosin method provided the best estimates of canopy cover in artificial images, whose accuracy was largely unaffected by variation in canopy density and object size. By contrast, LAB2 systematically overestimated canopy cover, because of the sensitivity of the method to small variations of chromaticity in artificial images. Winscanopy showed good performance when at least two regions per class were manually selected from a representative image. The results were replicated in real images of cultivated aromatic crops. The main findings indicate that digital photography is an effective method to obtain rapid, robust and reproducible measures of canopy cover in downward-looking images of agricultural crops, including aromatic plants.