Aplicação de algoritmos de aprendizado de máquina no reconhecimento de culturas e plantas daninhas

Abstract

Due to the technology advancement and the increase of its application in agriculture field, the use of solutions such as vision systems capable of distinguishing between crops and weeds even in earlier stages of development enables the management of these weeds before they cause the production loss. This work aims to apply and evaluate machine learning algorithms for crops and weeds classification. For this purpose, a dataset containing images from Danish agriculture was used and a set of characteristics related to the shape of these plants was extracted. In addition, the performance of the Support Vector Machine, Decision Tree and Logistic Regression classifiers were evaluated. The best result was obteined with the Support Vector Machine classifier whose accuracy and mean of the F-Measure was 82 %.