Data adversarial enhancement method without lucky ingredients for SAR image ship data

In response to the actual situation of the ship testing of SAR images, the data enhancement method has been widely proven to effectively improve the model effect. However, it is difficult to screen for effective data enhancement methods, but it is extensively enhanced. It is very likely that the model has been completed at the beginning of the model, but the weakness of the model itself has not been improved. For the characteristics of the SAR image ship dataset, this article proposes a new method of enhanced data enhancement for model weakness. This method is looking for a weak direction of the model to find a weak direction of the model. With each time from offense to data enhancement model optimization process, the performance of the model continues to improve.