Título: Data Fusion and Artificial Neural Network – Based Classification for Deforestation Monitoring in the Brazilian Amazon
Autores: Edson Costa Oliveira, Juan Lieber Marin & Gislan Silveira Santos
Resumo: The Amazon rainforest is widely recognized as one of the most biodiverse ecosystems on the planet, playing a fundamental role in global climate regulation. However, illegal deforestation continues to have profound international consequences, compromising Brazilian biodiversity and contributing to climate change. In this work, we present a classification approach based on artificial neural networks for analyzing spatial imagery and identifying deforested and non-deforested areas. The research incorporates geoprocessing methodologies, including the Linear Spectral Mixture Model, alongside advanced machine learning techniques such as Data Fusion. Experimental results demonstrate that both Multilayer Perceptron (MLP) and Convolutional Neural Networks (CNN) achieve high classification performance, with a ROC curve area exceeding 95%. In addition, data fusion techniques further improved the classification accuracy, enabling three of the developed models to correctly classify all samples presented to the network. These findings underscore the importance of continuing research on the Amazon biome and highlight the potential of artificial intelligence in supporting scientific investigations of this complex environmental system.
Palavras-chave: Artificial Neural Networks; Geoprocessing; Brazilian Amazon; Deforestation; Image Processing; Data Fusion.
Páginas: 8
Código DOI: 10.21528/CBIC2025-1174038
Artigo em PDF: CBIC_2025_paper1174038.pdf
Arquivo BibTeX:
CBIC_2025_1174038.bib