Data augmentation is one of the most important ways to improve object detection performance. In this paper, we propose a new offline data augmentation method, this method obtains multiple new anchors and new labels by fixed-point clipping in the original anchor box, by retaining important semantic information, the detection performance of the occlusion object is improved. At the same time, we produced a marine fish dataset with 11 kinds of marine creatures, and we verify the effectiveness of the proposed method in this dataset. Experiments confirmed the effectiveness of this method on the marine fish dataset, in the total dataset, the mAP@0.5 of this method is 10% higher than the Baseline. In the Tuna class that is seriously occluded in the obstruction problem, the mAP@0.5 of this method is 0.25 higher than baseline, which is increased by 46%. It indicates that this method can not only perform well in normal object detection but also maintain the advantage in complex scenes.