|
| 用于多维光学参数识别的钙钛矿人工光电突触 |
| Perovskite artificial optoelectronic synapses for recognition of multidimensional optical parameters |
| 投稿时间：2025-01-27 |
| DOI：10.3969/j.issn.1005-5630.202501270016 |
| 中文关键词: 钙钛矿 光电 人工突触 运动识别 |
| 英文关键词:perovskite optoelectronic artificial synapse motion recognition |
| 基金项目:国家自然科学基金面上项目(11974247) |
|
| 摘要点击次数: 0 |
| 全文下载次数: 1 |
| 中文摘要: |
| 人工智能视觉系统的发展很大程度上依赖于低延迟、高准确率的光学成像、检测和识别技术。传统的光电器件对于信息的处理速度有限，准确率较低，对实时光学信息以及信息中多维光学参数的处理能力有限。近年来，人工光电突触因其在光电转化中的记忆可塑性在人工视觉研究领域受到极大关注，但针对如何利用突触可塑性实现多维光学参数识别的关键问题尚缺乏研究。本文研发了一种基于钙钛矿薄膜的人工光电突触，实现了双脉冲易化特性和记忆可塑性。利用器件的突触可塑性进行了循环神经网络训练和人工神经网络训练，实现了对输入脉冲持续时间和激光功率90%以上的识别准确率，以及物体运动方向的识别准确率达100%。这些高准确率的结果为神经计算和机器视觉领域提供了一个新的解决方案。 |
| 英文摘要: |
| The development of artificial intelligence vision systems largely relies on the development of optical imaging, detection, and recognition technologies with low latency and high precision. Traditional optoelectronic devices have limited processing speed, low accuracy, and limited ability to handle real-time optical information and multidimensional optical parameters in the information. In recent years, artificial optoelectronic synapses have received great attention in the field of artificial vision due to their memory plasticity in optoelectronic conversion. However, there is still a lack of research on synaptic plasticity to achieve recognition of multidimensional optical parameters. This article develops an artificial optoelectronic synapse based on perovskite thin films, which achieves paired-pulse facilitation and memory plasticity. By utilizing the synaptic plasticity and the training based on recurrent neural network and artificial neural network, a recognition accuracy of over 90% for light pulse duration/power and an accuracy of 100% for object moving direction can be achieved. These results provide a new solution for neural computing and machine vision fields. |
| HTML 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
