MACHINE LEARNING TECHNIQUES FOR SELECTIVE LOGGING DETECTION IN X-BAND SAR IMAGES - A COMPARATIVE EVALUATION

Land use, land use changes and forest degradation have historically been the sectors that most contribute to greenhouse gas emissions in Brazil, according to the System of Estimates of Greenhouse Gas Emissions and Removals (SEEG), degradation being the major contribution. Therefore, the necessary containment of the increase in emissions is closely related to the control and combat of deforestation and forest degradation. Given the contribution of logging activity in this scenario, monitoring it is an important part of ensuring that the market is selling only products originating from sustainable exploitation. This work compares three methodologies, based on machine learning, for the detection of selective logging in X-band SAR satellite images collected over the Amazon, a tropical region with persistent cloud cover throughout the year, justifying the use of SAR data for monitoring. The Convolutional Neural Networks tested showed good performance in detecting gaps resulting from exploration, with the U-Net architecture showing the best result (accuracy 97\%) with the lowest pre-processing requirement, as it is a semantic segmentation approach, and not just classification, like the others.