Dynamic Model To Detect Weeds In Cereals Under Actual Fields Conditions

To implement spatially variable application of herbicides for weed control in arable crops, information on the distribution of weeds within the field is required. Manual surveying of weed patches is labor - intensive and not economic for production agriculture. As real-time patch spraying is still a difficult process, the overall objective of this research is to study the feasibility of automatically mapping weeds, a few days before the herbicide application.Taking into account that typical thresholding methods have shown some difficulties to cope with the complexity of cereal images, a segmentation method called dynamic model is proposed in this paper to distinguish between plants and soil. The method is derived from techniques known as 'embedded snakes', but in this case the model, represented by a binary image, evolves according to certain morphological transformations producing model contraction toward contour points in objects to be segmented (plants).It has been shown as the proposed technique improves segmentation results compared to other classical methods own to the better contour definition achieved. This is necessary in further steps devoted to shape discrimination between crop and weeds. On the other hand, the local characteristics of the algorithms permit a hardware implementation.