Dual‐Mode Optoelectronic Neuromorphic Architecture for Imaging‐Recognition of Chinese Characters

Abstract To overcome the intrinsic separation between sensing and computing in artificial intelligence vision systems, this study introduces a dual‐mode PbS/InP heterojunction photodetector that enables hardware‐level integration of photoelectric imaging and neuromorphic computing. In Mode 1, the device delivers a broadband spectral response (340–970 nm), high sensitivity (0.373 A W −1 at 870 nm), and precise 16 × 16‐pixel optical imaging of Chinese characters. In Mode 2, it reproduces synaptic behavior via photoelectric reservoir computing combined with a convolutional neural network, achieving 99.5% accuracy in Chinese radical classification and 97.1% in multi‐component character decomposition. Leveraging hardware system integration, a character‐component‐based recognition framework is developed that enhances pixel density for text recognition by 6.4‐fold, achieving 256 PPI. These results lay the foundation for multimodal artificial vision systems, with scalable potential through array expansion and integration with spiking neural networks.