A generative adversarial network based image augmentation method for ship segmentation in SAR images
Kai Zhao, Yan Zhou, Xin Chen, Huainian Zhang · 2020
Synthetic aperture radar (SAR) has good performance and is widely used. Ship monitoring plays an important role in the military field, and ship segmentation plays an important role in related research. SAR image segmentation with neural networks requires large amounts of data. In the case of a small data set, traditional data augmentation methods such as scaling and rotation methods have limited promotion on image segmentation. We propose a new method for SAR image augmentation: based on CycleGAN, combined with Slide Window Filtering, convert aerial images to SAR style images, and use it for SAR image segmentation. Experiments show that compared with the traditional data augmentation methods, this method can improve the segmentation result better.