Unregulated rates of deforestation within the Amazon rainforest have significantly impacted biodiversity, climate stability and other vital ecosystem services. To aid monitoring efforts, this study incorporated advanced Computer Vision (CV) techniques and moderate-resolution Sentinel-2 multispectral satellite imagery to detect areas of deforestation. To distinguish these areas, the U-Net, Residual U-Net, and Attention U-Net architectures were investigated for semantic segmentation. The results revealed that Residual U-Net and Attention U-Net models outperformed conventional U-Net models. Specifically, Residual U-Net achieved a validation F1-score of 0.95 and a test F1-score of 0.94, while Attention U-Net reached a validation F1-score of 0.95 and a test F1-score of 0.95. These results demonstrate the capabilities of advanced CV approaches and remotely sensed data in reliably detecting areas of deforestation. This work, potentially, contributes toward achieving the United Nations Sustainable Development Goals by providing a semi-autonomous framework to support effective conservation and restoration efforts in the Amazon.