光学合成孔径图像超分辨率重建技术研究

Research on Super-resolution Reconstruction Technology of Optical Synthetic Aperture Image

摘要:

针对光学合成孔径固有的中低频损失而导致的成像模糊问题,提出一种改进的超分辨率生成对抗网络(SRGAN)进行图像复原研究。首先通过MATLAB构建光学合成孔径图像数据集,并对数据集进行数据增强处理,其次根据ASPP网络设计思想,构建多尺度SRGAN生成器的残差结构,最后与传统超分辨率重建算法进行复原效果对比。实验结果表明,本算法可加快模型收敛速度,提升模型获取图像细粒度特征的能力,对于光学合成孔径图像的复原效果更优。

Abstract:

Aiming at the problem of image blurring caused by the inherent low-frequency loss of optical synthetic aperture, an improved super-resolution generative countermeasure net (SRGAN) is proposed for image restoration research. Firstly, the optical synthetic aperture image data set was constructed by MATLAB, and the data set was processed by data enhancement. Secondly, according to the design idea of ASPP network, the residual structure of multi-scale SRGAN generator was constructed. Finally, the restoration effect was compared with the traditional super-resolution reconstruction algorithms. Experimental results show that this algorithm can speed up the model convergence and improve the model"s ability to acquire fine-grained features of the image. It has a better restoration effect for optical synthetic aperture images.