粗糙面SAR图像渐进式生成对抗网络

Progressive Generative Adversarial Network for Rough Surface SAR Images

粗糙面合成孔径雷达(SAR)图像在遥感和目标识别等领域有着非常重要的意义。目前,粗糙面SAR图像的仿真方法主要有三种:数值法、统计法和解析法。数值法的计算复杂度会随着粗糙面尺寸的增大而升高,导致计算速度变慢,这限制了该方法的应用;统计法如空间相关模型是从统计角度生成SAR图像;解析法如基尔霍夫近似法(KA)等适用于计算粗糙面的散射矩阵。相干空变双向散射分布函数(CSVBSDF)物理模型可以生成多维度参数下的粗糙面的SAR图像,但其计算速度不能满足实时仿真需求。基于CSVBSDF,本文提出了一种粗糙面SAR图像渐进式生成对抗网络(RSPG),提高了SAR图像生成的速度。实验结果表明,生成的SAR图像与真实SAR图像在平均结构相似性指标上达到0.8,并且生成的速度与CSVBSDF相比得到了提高。

Rough surface synthetic aperture radar(SAR)images are of great significance in the fields of remote sensing and target recognition.In general,there are three types of methods for rough surface SAR imaging,i.e.,numerical,statistical,and analytical methods.The computational complexity of numerical method will increase with the increase in the rough surface size,which results in the increase in the computation amount and limits the application of this method.The statistical methods such as spatial correlation models simulate SAR images from a statistical perspective.The analytical methods such as the Kirchhoff approximation(KA)method are suitable for calculating the scattering matrix of rough surface.The physical model of coherent spatially varying bidirectional scattering distribution function(CSVBSDF)can generate rough surface SAR images of multi-dimensional parameters.However,it is too slow for real-time simulation.Based on the CSVBSDF,a progressive generative adversarial network for rough surface SAR images(RSPG)is proposed in this article,which speeds up the generation of SAR images.Experimental results show that the mean gradient structural similarity(MGSS)between the generated SAR images and the true images reaches 0.8,and the calculation speed is improved compared with the CSVBSDF.