This database contains images from Amazon and Atlantic Forest brazilian biomes used for training a fully convolutional neural network for the semantic segmentation of forested areas in images from the Sentinel-2 Level 2A Satellite. The images refer to the composition of bands 4, 3, 2 and 8. Each band was converted to a byte type (0-255). The images are still divided into three main sets: training, validation and testing: Training dataset: it contains 499 and 485 GeoTIFF images (Amazon and Atlantic Forest, respectively) with 512x512 pixels and associated PNG masks (forest indicated in white and background in black color). Validation dataset: it contains 100 GeoTIFF images for each biome with 512x512 pixels and associated PNG masks used for validation step. Test dataset: it contains 20 GeoTIFF images for each biome with 512x512 pixels for testing.