Automatic Aquatic Weed Classification Using Shape Analysis and Optimum-Path Forest

Abstract

The aquatic weed control through chemical substances has attracted much attention in the past years, mainly because of the ecological disorder caused by such plants with respect to the financial activities derived from the river, such as fishing and energy production, for instance. The fast growing of such plants may affect the whole ecosystem, degrading other essential plants that keep stable the local fauna and flora. Other negative aspect is the indiscriminate application of wrong chemical reagents, and also many times in inadequate quantities aiming the aquatic weed control, given that each specie has its own chemical control. Recent works have applied computer vision techniques aiming the identification of weeds in a certain region, for further application of chemical control in that river region. However, none study took into account the identification of the weed specie itself for further application of the correct chemical control, given that some of them are more effective than others for specific species. Therefore, the present research project has as the main goal a initial study of the automatic identification of three aquatic weed species, aiming a composition of a future expert system for the correct chemical control of such plants, using shape descriptors and supervised pattern recognition techniques, such as Optimum-Path Forest - OPF, Support Vector Machines - SVMs and Artificial Neural Networks - ANNs. It can be noticed that is the first time that OPF is used in this context, as well a comparison among supervised pattern recognition using shape features for the automatic recognition of aquatic weeds.