Single pixel image classification using ultrafast optoelectronic hardware

Smart cameras are a widespread technology for image recognition and robotic vision. However, these devices face bandwidth limitations related with the high number of pixels transmitted and analysed per image. Single Pixel Imaging (SPI) is an alternative imaging paradigm to conventional cameras offering unique advantages in resource-constrained environments. SPI can be combined with compressive sampling techniques to significantly enhance image acquisition efficiency and flexibility while reducing the system's complexity [1]. SPI has recently combined with digital light projectors based on chip-scale LED-on-CMOS technology [2] to achieve ultra-high image transmission rates [3]. In this study, we demonstrate an efficient SPI-based image classification system operating at kHz bandwidths. Furthermore, we study the performance of our SPI image classification approach for increasing levels of image compression, which correspond to higher transmission rates.