基于稀疏分解和预滤波处理的机载SAR图像超分辨率重建

Super⁃resolution reconstruction of airborne SAR images based on the sparse decomposition and pre⁃filtering processing

提升图像分辨率一直是合成孔径雷达(Synthetic Aperture Radar,SAR)成像处理的一个重要方向。该文对低分辨率SAR图像受相干斑噪声影响导致超分辨率重建性能下降的问题进行了研究,提出了一种基于字典稀疏表示和预滤波的超分辨率重建流程。对低分辨率雷达图像的稀疏表示可以实现图像不同结构成分信息的分解,然后采用复数神经网络实现模糊核的校正以及预滤波,从而提升图像质量以及低分辨率SAR图像与超分辨率网络的匹配程度。通过上述处理流程可以避免更改网络训练设置的复杂操作,达到改善重建图像的性能,同时抑制了超分辨率网络对相干斑噪声的放大。实验验证表明,加入低分辨率SAR图像的预滤波处理步骤后,超分辨率网络的重构结果在视觉效果及客观指标上都获得了明显提升。

Improving image resolution has always been an important direction of synthetic aperture radar(SAR)imaging processing.In this paper,the performance degradation of super⁃resolution reconstruction of low resolution SAR images caused by speckle noise is studied,and a super⁃resolution reconstruction flow based on dictionary sparse representation and pre filtering is proposed.Sparse representation is deployed to decompose the information of different structural components of low resolution radar images,and the complex valued convolutional neural networks(CNN)are used to correct the down⁃sampling kernels and conduct pre⁃filtering.Thus,the image quality and the matching degree between the low resolution SAR images and the super⁃resolution networks are improved.This process can avoid the complex operation of changing the network training settings,improve the performance of the reconstructed image,and restrain the super⁃resolution network from amplifying the speckle noise.The experimental results show that the reconstruction results of the super⁃resolution network can be significantly improved in terms of visual effect and objective indicators after introducing the pre⁃filtering process to the low resolution SAR images.