Super-coding resolution single-pixel imaging based on unpaired data-driven deep learning

Single-pixel imaging, which relies on structured illumination to restore two- or three-dimensional scenes from one-dimensional bucket signals, has been widely concerned. However, the resolution of the reconstructed image is consistent with that of the illuminated patterns. The need for higher resolution single-pixel imaging, on the one hand, can be achieved through the higher resolution illuminated patterns, but this means difficult spatial transmission and more measurements. On the other hand, it can be realized by means of image super-resolution algorithm, but the performance of existing algorithms is often not satisfactory. In this paper, a super-coding resolution single-pixel imaging method is proposed. Low-quality preliminary results with low pixel resolution are preferentially obtained. With the help of the unpaired training strategy of the Cycle-gan network, higher resolution, and higher quality image reconstruction can be achieved while reducing the pressure of the dataset and measurements. This method will help to improve the performance of single-pixel imaging and broaden its application range.