EFCformer: high-resolution image restoration network for optical synthetic aperture imaging system

Synthetic-aperture optical imaging systems use multiple sub aperture arrays to significantly improve the imaging resolution of space telescopes. However, the sub aperture arrangement inevitably impacts the intermediate and low frequencies of the modulation transfer function, which results in blurred images. This study led to the proposal of a dual-feature extraction network based on convolution and a transformer, to effectively recover high-resolution images from synthetic aperture optical systems. Specifically, the proposed network consists of a new convolution layer for local feature extraction and a new transformer layer that focuses on global information. The introduction of concentrated linear attention and a newly developed gated forward propagation module enables the computational load of the transformer to be reduced to ultimately improve the ability to extract global information. To avoid any adverse effects by the ringing phenomenon generated in the synthetic aperture optical imaging system during image restoration, we used a new feature enhancement fusion module to combine the extracted features of the convolution and transformer layers and enhance them to strengthen the ability to sharpen the expression of the structural features. The experimental results indicated that, compared with other advanced methods, our method can improve the peak signal-to-noise ratio by 1.5% and reduce the number of model parameters by 17% to effectively restore the high-resolution image of the synthetic aperture optical system.