Uso de redes neurais artificiais para determinação de pixels indicando água em uma imagem de satélite

Abstract

Floods are problems faced by countless people in different locations. Filling can be defined as a large amount of water flowing due to excessive rainfall. The process of measuring and predicting verified areas is often a costly process, whether in terms of knowledge or hardware for map processing, using classic methods of remote sensing. Knowing this, this work brings an alternative to the use of classical methods, where it is proposed to classify and identify the flooded areas, a model based on Artificial Neural Network (ANN). The use of Artificial Neural Networks today is being widespread and the great advantage of using ANNs is their great adaptability. By using an ANN-based model, it is possible, for example, to train an ANN to detect pixels that indicate water in the Hydrographic Basin of a certain river, as the standard in this work, and later adapt and train this model for other Hydrographic Basins. Therefore, when using an ANN model based on a Multi-Layer Perceptron (MLP), which is a model specialized in data classification, it can be concluded that image segmentation can be performed through it, but that there are limits. Four models based on MLP were implemented, which received training based on extracted data of 7 bands of the OLI sensor of the Landsat 8 satellite. After training, the four models generated masks in black and white, indicating when in black that the pixel was classified as being “water”, otherwise as “not water”.