Robust Methods for Measurement of Leaf-Cover Area and Biomass from Image Data

Abstract

Leaf-cover area is a widely required plant development parameter for predictive models of weed growth and competition. Its assessment is performed either manually, which is labor intensive, or via visual inspection, which provides biased results. In contrast, digital image processing enables a high level of automation, thereby offering an attractive means for estimating vegetative leaf-cover area. Nonetheless, image-driven analysis is greatly affected by illumination conditions and camera position at the time of imaging and therefore may introduce bias into the analysis. Addressing both of these factors, this paper proposes an image-based model for leaf-cover area and biomass measurements. The proposed model transforms color images into an illumination-invariant representation, thus facilitating accurate leaf-cover detection under varying light conditions. To eliminate the need for fixed camera position, images are transformed into an object–space reference frame, enabling measurement in absolute metric units. Application of the proposed model shows stability in leaf-cover detection and measurement irrespective of camera position and external illumination conditions. When tested on purple nutsedge, one of the world's most troublesome weeds, a linear relation between measured leaf-cover area and plant biomass was obtained regardless of plant developmental stage. Data on the expansion of purple nutsedge leaf-cover area is essential for modeling its spatial growth. The proposed model offers the possibility of acquiring reliable and accurate biological data from digital images without extensive photogrammetric or image-processing expertise.