光学合成孔径成像系统中光学传递函数的频率响应下降,会不可避免地导致成像模糊,因此通常需要借助维纳滤波或盲解卷积算法来实现图像复原,最终获得清晰的高分辨率图像。提出一种基于U型卷积神经网络的深度学习框架,通过MATLAB软件构建数据集,以对网络进行训练,并将所训练的U型网络与盲解卷积算法的图像复原效果进行对比。数值仿真结果表明,在弱噪声条件下,U型网络在基于光学合成孔径成像系统的图像复原中展现出较强的复原能力以及一定的泛化能力和通用性,能够实现图像的快速盲复原,因而具有潜在的应用前景。

The decrease of the frequency response of the optical transfer function in the optical synthetic aperture imaging system will inevitably lead to image blur. Therefore, it is usually necessary to use the Wiener filtering or blind deconvolution algorithm to achieve image restoration, and clear and high-resolution images are obtained finally. A deep learning frame based on a U-shaped convolutional neural network is proposed. The data set is constructed by the MATLAB software to train the network. The image restoration effects of the trained U-shaped network and blind deconvolution algorithm are compared. The numerical simulation results show that the U-shaped network has strong recovery ability, generalization ability, and versatility in the image restoration based on the optical synthetic aperture imaging system under the condition of weak noise. It can realize fast blind restoration for images and has potential application prospects.