Aquatic weed automatic classification using machine learning techniques

Aquatic weed control through chemical products has attracted much attention in the last years, mainly because of the ecological disorder caused by such plants, and also the consequences to the economical activities. However, this kind of control has been carried out in a non-automatic way by technicians, and may be a not healthy policy, since each species may react differently to the same herbicide. Thus, this work proposes the automatic identification of some species by means of supervised pattern recognition techniques and shape descriptors in order to compose a nearby future expert system for automatic application of the correct herbicide. Experiments using some state-of-the-art techniques have shown the robustness of the employed pattern recognition techniques.

► A novel system for automatic aquatic weed recognition. ► To be the first to apply the optimum-path forest classifier for such context. ► We provide a dataset composed by images of three species of aquatic weeds. ► A wide comparison among several state-of-the-art pattern recognition techniques.