Método baseado em aprendizagem de máquina para detecção de pequenas porções d’água em imagens multiespectrais adquiridas por drones

Abstract

RPAS − Remotely Piloted Aircraft System, popularly known as drones, are used to automatically identify objects and scenarios (normally water tanks, buckets, plant pots, and other containers contained in open-air trash) that characterize potential mosquito breeders, such as Aedes aegypt, from the acquired images. However, despite knowing that water stagnation is an essential condition for mosquito breeding, computer vision systems (SVCs) proposed in the literature for automatic image analysis do not include the detection of water in suspicious objects and scenarios, which which constitutes a technical limitation for the effective use of drones in vector monitoring and control actions. In this work, a method is proposed that employs an Artificial Neural Network of the Multilayer Perceptron type (RNA−MLP) to identify small portions of water in multispectral images acquired by drones. To carry out the experiments, a database was composed containing 151 multispectral images of 1280×960 pixels, which are divided into two sets and were acquired from simulated scenarios containing small containers with and without water in a controlled environment. The results obtained by the proposed method (average precision of 0.759) corroborate its potential to increase the technical solutions of existing SVCs, making them more effective in combating mosquito outbreaks, which can bring contributions to the area of public health.