用激光神经形态芯片实现低功耗粒子分类，速度0.01-0.1米/秒
A VCSEL based Photonic Neuromorphic Processor for Event-Based Imaging Flow Cytometry Applications
- 基于VCSEL的光子神经形态系统，将图像转为1比特脉冲流
- 95.8%分类准确率，最多96个脉冲，压缩率达98.4%-99.5%
- 适合资源受限场景，如便携式细胞检测设备
本文提出一种基于可激发VCSEL的时间延迟（TD）极限学习机与事件驱动2D相机集成的光学神经形态成像处理细胞分析系统。该系统用于分类直径不同的聚甲基丙烯酸甲酯（PMMA）颗粒，颗粒运动速度在0.01至0.1米/秒之间。所提出的光子方案实现了95.8%的分类准确率，将原始2D图像编码为最多含96个脉冲的1比特脉冲流。此外，合成帧的二值化表示使存储需求降低98.4%至99.5%，硬件需求降低50%至84%。这些结果表明，神经形态计算与传感的融合可推动低功耗、低延迟应用在资源受限环境中的发展。
原文摘要 · Abstract (English)
This work presents an optical neuromorphic imaging and processing cytometry system that integrates an excitable VCSEL-based time-delayed (TD) extreme learning machine with an event-based 2D camera. The proposed system is designed for the classification of Polymethyl Methacrylate (PMMA) particles of varying diameters moving at speeds between 0.01 and 0.1 m/s. The TD photonic scheme achieved a classification accuracy of 95.8% while encoding the original 2D images into a 1-bit spike stream containing a maximum of 96 spikes. Additionally, the binary representation of the synthetic frames enables a significant reduction in memory and hardware requirements, ranging from 98.4% to 99.5% and 50% to 84%, respectively. These findings demonstrate that the integration of neuromorphic computing with sensing can facilitate the development of low-power, low-latency applications optimized for resource-constrained environments
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。