High-quality single-pixel imaging via untrained attention network at a low sampling ratio

An innovative single-pixel imaging (SPI) method based on untrained attention network (UAnet) is proposed. The SPI technique illuminates the target with a sequence of modulated light fields. And a single-pixel detector (SPD) is used to collect the light intensities. The image is obtained through reconstruction algorithm combining the light fields and intensities. In the novel method, we incorporate the attention gate and SPI model into the untrained Unet in order to achieve high-quality imaging at a low sampling ratio. The untrained Unet has the advantage of good generalization ability without pre-training. The attention gate can efficiently extract the main features of the target. Numerical simulations and experiments demonstrate the UAnet can obtain better image quality at a low sampling ratio (less than 10%) than the other existing algorithms. This method effectively improves the imaging quality and efficiency of SPI.