Single-Pixel Image Reconstruction and Super-Resolution using Deep Learning

Single-pixel imaging (SPI) utilizes a single-point detector and modulated illumination patterns to capture images through measurement acquisition and reconstruction. However, existing SPI reconstruction techniques are inefficient due to their iterative nature and high-sparsity sampling patterns. This research develops a deep learning-based model and a novel SPI basis to enhance the imaging efficiency and accuracy of SPI while minimizing computational demands. Experiments show that the proposed super-resolved SPI system can achieve significantly higher imaging resolution and improved accuracy at low sampling ratios in real time and can be readily integrated into more advanced SPI technologies in the future.