PREDIZENDO A DEPOSIÇÃO DE PRODUTOS FITOFARMACÊUTICOS COM APRENDIZADO DE MÁQUINA

Abstract

Population growth poses a number of challenges, including increasing food production and supply. To overcome such an obstacle, large-scale agricultural production is essential. In addition, agriculture is a strategic area in the economy of several countries, driven by increased production. In this scenario, a common practice is the management of protection with the spraying of plant protection products in order to protect the crop from the harmful action of agricultural pests. During this activity, it is common for the product to drift out of the target region, contaminating environmental regions or causing excessive deposition in agricultural subareas due to overlapping of the product. The possibility of the sprayer element adapting to the weather conditions is seen as an alternative to reduce the drift of the sprayed product, increasing its accuracy in the deposition of the product and, consequently, providing a more suitable environment for the growing crop. For this, it is necessary that the sprayer element can perform the prediction of the deposition of the product at run time. However, the current approaches have a high computational cost, causing this problem to require an exaggerated time for its processing. In this sense, this work proposes a system as proof of concept for a new approach based on machine learning for the prediction of the pulverized product in order to represent in a similar way the real behavior of the deposition. The techniques of artificial neural network, regression tree, support vector machines and random forests were deployed with different types of parameter settings. The results show that this approach allows to achieve a satisfactory representation of the actual deposition and stimulates new studies in order to compare the computational cost of the proposed approach with other existing ones.