Abstract: Scene classification in optical remote sensing (RS) imagery has been extensively investigated using both learning-from-scratch approaches and fine-tuning of ImageNet pretrained models.