Research on single-pixel imaging reconstruction algorithm based on residual attention channel network

Single-pixel imaging (SPI) refers to the complete image obtained by processing and reconstructing the data acquired from the image, however, the traditional SPI reconstruction algorithm reconstructs the image with more detail texture missing and serious artifacts at low sampling rates. To address this problem, this paper proposes an SPI reconstruction algorithm based on residual in-channel attention block (RICAB-GAN). In the proposed reconstruction algorithm, we improve the residual block in the generator network by adding channel attention (CA), which enables the network to notice more texture details in the image. In the discriminator network we remove all the BN layers to reduce the artifacts in the image and improve the generalization ability of the model. Through simulation experiments, our algorithm improves psnr by 1-5dB, ssim by 0.1-0.3, and retains image details and textures better than traditional SPI algorithms at low sampling rates.