神经形态工程学
记忆电阻器
计算机科学
卷积神经网络
材料科学
特征(语言学)
人工智能
特征提取
Boosting(机器学习)
人工神经网络
计算机视觉
加密
电子工程
运动检测
目标检测
运动估计
机器视觉
运动控制
执行机构
光子学
等离子体子
激发
光电子学
功能(生物学)
模式识别(心理学)
光学计算
作者
Yongxing Zhu,Ye Tao,Zi Wang,Jingyao Bian,Xiaoning Zhao,Ya Lin,Haiyang Xu,Ying Liu
摘要
ABSTRACT All‐optically controlled polarization‐sensitive memristors offer a promising route toward compact and efficient neuromorphic vision hardware by unifying polarized light sensing and synaptic computing within a single device. However, most reported polarization‐sensitive neuromorphic systems still rely on electrical signals for synaptic modulation, limiting true all‐optical operation and increasing system complexity. Here, we propose an all‐optically controlled polarization‐sensitive memristor based on an Au nanoparticle/WSe 2 nanobelt heterojunction, which serves as a unified platform for motion feature extraction. The device exhibits not only angle‐dependent synaptic behavior but also polarization‐sensitive, reversible, and fully optical synaptic modulation, attributed to the surface plasmon resonance (SPR) of Au nanoparticles and the optical excitation of WSe 2 . Utilizing all‐optical polarization‐dependent synaptic plasticity, dynamic encryption of image information is achieved using 532 and 320 nm laser illumination. Furthermore, polarization‐state‐dependent convolutional kernels allow effective extraction of motion‐related features, leading to enhanced handwriting recognition accuracy under noisy conditions and enabling polarization‐assisted moving object detection. This work demonstrates that all‐optically controlled polarization‐sensitive memristors can function as a unified hardware platform for motion feature extraction, paving the way for advanced polarization‐sensitive neuromorphic vision systems.
科研通智能强力驱动
Strongly Powered by AbleSci AI