Machine-Learning-Enhanced Intelligent Recognition of Integrated Neuromorphic Vision Sensors Based on Copolyurethane

Artificial photoreceptors utilizing piezoelectric polymers and semiconductors can convert external mechanical deformations, forces, or changes in light into electrical signals, making them essential for advanced optoelectronic sensors and smart wearable devices. However, this approach faces several challenges, including slow response time, weak signal, and high power consumption. This study synthesizes a series of polyurethanes containing azobenzene-based photoisomer units and ionic-liquid-based dipole units (comprising loose cation-anion pairs) based on the nanophotoelectric effect, wherein ultraviolet light induces isomerization of photoisomer segments and generates dynamic dipoles, creating equal amounts of charges with opposite signs at the electrodes. The nanophotoelectric generator achieves open-circuit voltage of 37 V, short-circuit current of 265 µA, and rapid response time of 7.5 µs under UV illumination. Furthermore, 81 individual nanophotoelectric generators are integrated into a 9 × 9 pixel array for a machine-learning-assisted system to accurately (96.22%) recognize different items, like human vision; it simultaneously executes super-resolution refinement on the acquired pixel images, further improving the identification results. Precise, efficient intelligent object recognition is thus attained through material innovation, and a comprehensive system is established that encompasses azobenzene-ionic-liquid copolymer preparation, device assembly, integration, signal acquisition, and machine learning, offering novel insights into bionic visual recognition systems.