Extracting fuzzy rules to describe weed infestations in terms of soil factors

Site-specific treatment of weeds in precision agriculture is essential to know which factors determine a high occurrence of weeds. Former statistical studies searching for relationships between individual soil variables and wild-oat occurrence found no clear linear relationship among the analyzed variables. However, farmers observations pointed to better wild oats grow in specific locations where the physical and chemical conditions are probably more conducive. Data mining is justified in applications where conventional analysis methods are not able to extract useful and task-oriented knowledge. This paper proposes a fuzzy rule-based model, generated from a supervised learning process, to describe the complex relationships among weed occurrence and some soil properties. A genetic algorithm is used to derive the set of fuzzy rules by driving a search in the space of possible solutions (or models). This linguistic model has the advantage of being both intuitive and directly interpretable by the operator involved in the field tasks.