DETECÇÃO DE OBJETOS EM PLANTAÇÃO DE EUCALIPTO

Abstract

The objective of the work is to look among some selected algorithms, which obtains the best performance in detecting termites in a eucalyptus plantation, using average precision (AP) and recall, trying to prevent it from being overfit, so that it can be used in other datasets of the same object. Object detection has achieved incredible performances in recent years, with the emergence of new detection algorithms and methods, it has become increasingly easy to training a machine to identify objects. With the set of Images properly prepared for the training of an AI, you can detect a diversity of objects at the same time. Giving an algorithm an image is expected to return the same image to us with one or more possible results for the searched object, which can be identified even if with some kind of visual obstruction. With object detection being used by a lot of companies from different areas, algorithms being updated and getting new variations and becoming more accessible, the proposal is to train new and old algorithms, with the objective to find an algorithm that can adapt to a farm ambient and find termites, analyze their results and try to find which characteristics makes an algorithm to perform better.