Classificação de imagens de satélite com redes neurais convolucionais

Abstract

Satellite images are used in several areas, such as agribusiness, urban planning, environmental monitoring, among others. These images have a large and complex amount of data and the application of Machine Learning techniques can help in the automatic classification of the information contained in them. This article aims at the theoretical and practical development of a convolutional neural network that can classify satellite images taken from the Amazon Basin, thus being able to be used as a tool for environmental monitoring and detection of changes in the environment. To carry out this work, the Planet dataset: Understanding The Amazon from Space was used. This dataset has 40,479 images for model training and 40,479 images for testing, divided into 17 distinct categories that encompass atmospheric conditions and land use/land cover. At the end of the article, the reader will have an understanding of the theory behind convolutional neural networks from the practical presentation of a model developed and with accuracy greater than 95% in the classification of satellite images.