Validation of a virtual agronomic image modelling

In Precision Agriculture, remote sensors such as imaging systems are commonly used to manage site-specific weeds. Although there are several algorithms using spatial information, few tests proving their efficiency have been performed on relevant database. In this paper we propose an innovative approach based on a simple model that simulates virtual crops to evaluate the robustness and efficiency of any crop/weed discrimination algorithm. A two-dimensional statistical and spatial crop field model is presented: it allows the design of virtual crops with given characteristics (plant size, inter-row width…) for various species (wheat, sunflower, maize…). The Weed Infestation Rate is set on demand at three different spatial distributions (point, aggregate, or a mixture) providing the diversity of the resulting simulated field. The spatial realism of the model is validated by two spatial descriptors (nearest neighbor method and Besag's function) at different scales through a pair of real/virtual pictures.