High-Quality Single-Pixel Imaging Based on Large-Kernel Convolution under Low-Sampling Conditions

In recent years, deep learning has been introduced into the field of Single-pixel imaging (SPI), garnering significant attention. However, conventional networks still exhibit limitations in preserving image details. To address this issue, we integrate Large Kernel Convolution (LKconv) into the U-Net framework, proposing an enhanced network structure named U-LKconv network, which significantly enhances the capability to recover image details even under low sampling conditions. Compared to conventional deep learning networks, the U-LKconv network can reconstruct images with higher Signal-to-noise ratio (SNR) and more detailed features at the same sampling rate. Specifically, at a sampling rate of 4.8%, our method achieves peak performance, with a PSNR of 28.41 dB and an SSIM of 0.852 during the tests, surpassing the performance of other comparison methods, the superiority of the proposed method is also validated in experiment. Additionally, our approach demonstrates a faster convergence rate compared to other deep learning networks, requiring only 30 epochs for network convergence. Consequently, U-LKconv network can be addressed as an exceptionally advantageous solution, not only in terms of image fidelity and robustness but also due to its significant computational efficiency, which makes it highly applicable in the field of low-sampling SPI and drives the practicalization of single-pixel imaging.