Machine learning-based deforestation detection in the Amazon Rainforest

a data-driven comparative study

Resumo

The preservation of the Amazon Rainforest, a crucial ecosystem for global climate regulation, requires effective and systematic monitoring of deforestation activities. In this context, machine learning (ML) algorithms offer robust, well-established tools for detecting and mapping deforestation occurrences from remote sensing (RS) imagery. In fact, by combining advanced ML-based techniques with RS technologies, one can monitor and analyze vast amounts of forest areas, identifying potential changes in land cover over time while still facilitating conservation measures and sustainable land management practices. Considering the above context, in this work we conduct a comparative analysis of three machine learning-based frameworks: Isolation Forest (IF), One-Class Support Vector Machine (OC-SVM) and Convolutional Neural Network (CNN). More precisely, by integrating remote sensing data collections, image processing techniques, and machine learning algorithms, we adapt and refine the aforementioned machine intelligent approaches for the task of deforestation detection in the municipality of São Félix do Xingu, Pará, Brazil: one of the largest municipalities in the world in terms of native land area.