Our aim is to propose a probabilistic characterization and generation of agricultural surfaces by structuring objects (aggregates, clods and wholes) from high resolution images. We propose to describe agricultural surfaces by two levels of roughness: The first one corresponding to clods, aggregates and holes and, the second being the substrate on which are set down these objects. Having a segmentation algorithm by Hierarchy of Contours (HC) for objects identification, we have highlighted the influence of the gradient estimation method on that algorithm. We have also adapted a mathematical morphology approach - Watershed - for objects identification. To improve the boundaries of the detected objects, we developed an algorithm based on simulated annealing algorithm to move the clod boundaries. We show that semi-ellipsoid is the mathematical shape which model well objects. Having estimated the probability lows of the semi-ellipsoid parameters (orientation, major and minor axis, high), and studied their dependence, we developed a procedure to generate objects on a plane surface. We show that generated objects have the same statistics that identified objects on the high resolution images. We show that isotropy of the surface is related to orientation of the objects and that there exists a high correlation between substrate and objects placed on the plane.
