基于三氧化钨光致变色材料的全光视觉感知学习
All Optical Visual Perceptual Learning Based on Tungsten Trioxide Photochromic Materials

人类的视觉系统承担了大部分的信息感知任务，视网膜检测含有信息的入射光，如强度、波长和持续时间，将其转换为神经脉冲，并通过光神经和突触将其传递给大脑。模拟视觉感知的功能将是迈向机器人视觉和人工智能的关键一步。与电刺激的人工突触相比，光电突触因其高带宽，低能耗，高传输速度等特点在构建神经网络中有巨大潜力，但由于光电突触在读取信息时仍要进行光电转化，使得降低能耗、提高速度存在一定瓶颈。因此，我们基于三氧化钨光致变色材料进行全光的视觉感知学习模拟，零接触读取光信号，验证了材料的学习经验行为，并通过人工神经网络证明了延长对材料的训练时间可以提高对手写数字的识别精度，为实现构建更快、更省能的全光视觉感知神经系统提供了新思路。
The human visual system undertakes most of the information-sensing task, with the retina detecting incident light containing information, such as intensity, wavelength, and duration, converting it into nerve impulses, which are transmitted to the brain through nerves and synapses. The ability to simulate visual perception will be a key step toward robotic vision and arti-ficial intelligence. Compared with electrically stimulated artificial synapses, photoelectric synapses have great potential in the construction of neural networks due to their characteristics of high bandwidth, low energy consumption, and high transmission speed. However, photoelectric synapses still need to be converted when reading information, so there is a certain bottleneck in reducing energy consumption and improving speed. Therefore, we simulated all-light visual perception learning based on tungsten trioxide photochromic materials, and read optical signals with zero contact, which verified the learning experience behavior of materials. Moreover, through artificial neural networks, we proved that extending the training time of materials can improve the recogni-tion accuracy of handwritten digits, providing a new idea for realizing the construction of a faster and more energy-saving visual perception nervous system.