基于神经网络的SAR图像超分辨率重建技术研究

Research on super-resolution reconstruction technique of SAR images based on neural network

合成孔径雷达(SAR)图像在地形测绘、农作物监测等领域有重要作用。为改善SAR图像分辨率,本研究利用基于生成对抗网络(SRGAN)支持下的SAR图像超分辨率重建方式,改进模型加载数据结构,使用同一区域的哨兵一号(Sentinel-1A)雷达卫星SAR影像和高分三号卫星SAR影像形成训练模型的数据集,将哨兵一号雷达卫星SAR图像的地物细节提高到接近高分三号卫星SAR图像数据的级别。实验表明,该方法能提升极化方式为VV的哨兵一号雷达卫星SAR图像的地物细节。

Synthetic aperture radar(SAR)images play an important role in the fields of terrain mapping and crop phenology monitoring etc.In order to improve SAR image resolution,using the super-resolution reconstruction of SAR images supported by SRGAN,the model loading data structure is improved by use of Sentinel-1A radar satellite SAR images and GF-3 satellite SAR images of the same region to form the data set of training model,with the feature detail of Sentinel-1A radar satellite SAR images improved up to a level close to that of GF-3 satellite SAR image.The experiments show this method can enhance the feature detail of Sentinel-1A SAR images with the polarisation mode of VV.