Resumo:
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.
