Redução de uso de agrotóxicos por meio de inteligência artificial

Abstract

The cultivation of sugarcane has gained great prominence in several countries due to its large diversity of application. The modernization of agriculture has allowed high productivity in several cultivars, which are largely affected by the invasion of weeds. With sustainable agriculture, the use of herbicides has been strongly avoided, and more effective pest control methods have been required. In this survey, several artificial intelligence models capable of identifying the invasion of weeds in a sugarcane cultivar were proposed. In addition, four colors spectra based on a multispectral camera coupled to an unmanned aerial vehicle were employed as input variables. The models studied in this work were: GAMLSS, Neural Networks, Decision Trees, Random Forests, Bagging Cart, Gradient Boosting Machine, KNearest Neighbors and C4.5. Among the studied models, the Random Forests presented the best performance. For each model, all computational codes was presented for future new surveys. The precise identification of the weed infestation place allowed field management to be carried out only by the herbicide application in the damaged points. This procedure avoids an increase in the production costs, and mainly prevents the soil contamination and generation of hazardous effluents. Thus, avoiding an increase in production costs and, above all, it is avoiding unnecessary contamination of soils and effluents.