MINING SITES RECOGNITION VIA SCENE CLASSIFICATION USING 3D CONVOLUTIONAL NEURAL NETWORKS

High spatio-temporal resolution satellite imagery has become increasingly accessible, thanks to advancements in the aerospace industry which, combined with a growing computational power, has enabled the spring of novel techniques regarding recognition in remote sensing (RS) images. That being said, there is still a lack of studies around 3D convolutions for spatio-temporal data applied to classification problems in RS. Hence, this study investigates the application of 3D convolutional neural networks (3DCNNs) for scene classification in RS images, focusing on the identification of mining sites for its  high environmental and social impact. We firstly developed a dataset based on MapBiomas products and Planet imagery, then we evaluated the effectiveness of 3DCNNs in capturing temporal information from a sequence of monthly captured images. Comparison between a 3DCNN and a traditional approach, using only a single image with a 2DCNN, shows the advantage of the 3DCNN solution with NIR-G-B images.