Leveraging background augmentations to encourage semantic focus in self-supervised contrastive learning