Single-pixel imaging using deep learning, FPGA, and non-negative matrix factorization

Single-pixel imaging is a unique technique that can image target objects over a wide band range and in low-light environments. However, this method requires a large number of structured pattern illuminations in addition to reconstruction calculations, which limits its practical application. We introduce the single-pixel techniques that we have developed: deep learning, optimization, field-programmable gate array (FPGA), and nonnegative matrix factorization (NMF)-based methods. Deep learning and optimization techniques can improve image quality, and FPGA approaches can perform real-time imaging. The NMF approach can drastically reduce the number of structured patterns and measurement time.