This study introduces a novel attention mechanism for object detection in optical remote sensing images. In aerial imagery, many small objects rely on environmental cues for recognition. Their successful recognition largely depends on their surrounding environmental backdrop since the environmental information can provide crucial clues about the object's shape, orientation, and other features. Therefore, we firmly believe in the enormous potential of the Swin Transformer in capturing these prior knowledge. Our strategy, based on RepPoints and using Swin Transformer as a backbone, engages in sample selection through the fusion of Shape-Adaptive Selection and Shape-Adaptive Measurement strategies. This strategy fully exploits the shape information of the targets, dynamically optimizing the process of positive and negative sample matching. We have also designed a boundary center loss function to enhance localization precision and incorporated an orientation classification prediction head to explicitly predict target angles. Experiments on a private SAR dataset and DOTA benchmark validate the superiority of our approach in improving accuracy, efficiency and reducing model size compared to other methods. This research paves the way for new directions in Transformer-based rotational object detection.