Abstract: Traditional neuromorphic devices used for pattern recognition are mostly based on electrical signals, which suffer from bottleneck problems such as bandwidth limitation, resistance capacitance delay, and current crosstalk, leading to image distortion and seriously affecting pattern recognition results. Therefore, it is crucial to avoid signal crosstalk caused by electrical signals to ensure the effectiveness of subsequent pattern recognition results. Compared with electrical signals, optical signals have the advantages of fast processing speed, strong anti-interference and small energy loss. This article proposes a single-layer all-optical device based on phosphorescent materials, which successfully simulates the basic functions such as excitatory postsynaptic intensity (EPSI) and paired-pulse facilitation (PPF), which are similar to biological synapses. The device can process visual information and its luminescent properties are similar to synapses, achieving diversified utilization of light in information display and recognition. Finally, the MNIST dataset was used to recognize handwritten digits, achieving a pattern recognition accuracy of 97.36%.