Identification of weeds in sugarcane fields through images taken by UAV and Random Forest classifier

Sugarcane is one of the most important cultures in the world. The productivity of sugarcane is affected by many factors, among them weeds can cause several problems. Weed control is made usually by herbicides application because sugarcane occupies extensive areas, and due to the same reason, the decision about herbicide type and dosage has been done by sampling. This work mode does not allow variation and causes problems of herbicide application, since the presence and weed type may not be uniform in whole field. There are some solutions based on satellite image analysis that allow the coverage of the entire field, solving the problem caused by sampling sense, but this solution depends on high weed infestation and a clear sky for good results. This work proposes a system for weed surveying, based on image pattern recognition with pictures taken by a UAV (Unmanned Aerial Vehicle); this alternative can take pictures very close to the plants, which allows species recognition in lower infestation levels and without clouds interference. This solution achieved an overall accuracy of 82 % and kappa coefficient of 0.73 in preliminary tests.