MODICOVI - Monocot and dicot coverage ratio vision based method for real time estimation of canopy coverage ratio between cereal and dicotyledon weeds Henrik S. Midtiby 1* , Morten S. Laursen 1 , Rasmus N. Jørgensen 1 and Norbert Krüger 1 1 University of Southern Denmark, Campusvej 55, 5230 Odense M, Denmark. * Corresponding author. E-mail: hemi@kbm.sdu.dk Abstract In agriculture, weed control is becoming an increasing problem due to limitations on the amount and on the number of herbicides permitted. If herbicide application could be targeted areas containing weed infestations the used amount of herbicide could be significantly reduced and still achieve efficient weed control. This paper presents a real time and leaf occlusion robust method, Modicovi, capable of estimating the mono- and dicotyledon leaf coverage ratio within a given imaging area. Modicovi is based on locating vegetation boundaries in the input image and investigates the relative location and orientation of nearby edges. The scope is limited to handle maize plants up to the 2 leaf stage where monocotyledons will have oblong narrow leaves whereas dicotyledons will have leaves with a circular shape. This work is based on these central differences in shape to estimate the monocotyledon/dicotyledon ratio. Keywords: perimeter analysis, weed density estimation 1. Introduction Modern agriculture relies on efficient weed control, of which automation is of high interest as described in (Slaughter, D. et al. 2008). A central part is to estimate the weed pressure in acquired images. Usually two approaches are used 1.) using a set of morphological features to describe the shape for example as done by Weis, M. et al. 2010 or using a shape model derived from the statistics of recorded images as performed by Søgaard, H. 2005. Both of these methods has problems with partially occluded leaves. The only promising work known to the author in the case of partial occlusion is Kaspersen, K. et al. 2010. The aim of this work is to give a generalized measure of the weed to crop ratio for patches of 50x30cm in real time (delay less than 60ms from beginning of image exposure to activation of nozzle). In this paper an algorithm named Modicovi (MOnocot and DIcot Coverage ratio VIsion) which estimates the monocot vs. dicot ratio by analyzing the perimeter of the leaves is described. The scope of the algorithm is limited to the process of how to reach from a segmented (binary) image to a description of the weed pressure.