基于Gabor滤波器的事件流特征增强及事件相机对象识别  

Gabor Filter-based Event Stream Feature Enhancement and Event Camera Object Recognition

作　　者：周茜[1,2,3] 郑鹏 ZHOU Qian;ZHENG Peng(Hebei University of Technology,State Key Laboratory of Reliability and Intelligence of Electrical Equipment;School of Health Sciences and Biomedical Engineering,Hebei University of Technology,Hebei Key Laboratory of Bioelectromagnetics and Neural Engineering;School of Health Sciences and Biomedical Engineering,Hebei University of Technology,Tianjin Key Laboratory of Bioelectricity and Intelligent Health)

机构地区：[1]河北工业大学,省部共建电工装备可靠性与智能化国家重点实验室 [2]河北工业大学生命科学与健康工程学院,河北省生物电磁与神经工程重点实验室 [3]河北工业大学生命科学与健康工程学院,天津市生物电工与智能健康重点实验室

出　　处：《仪表技术与传感器》2024年第4期76-80,共5页Instrument Technique and Sensor

摘　　要：基于Gabor滤波器的事件驱动卷积是仿生分层脉冲神经网络中常用的事件相机对象特征提取方法。为提高该类网络事件相机对象特征提取能力,提出基于Gabor滤波器的事件流特征增强算法,并应用于奖励调节STDP规则的脉冲神经网络事件相机对象识别系统。算法首先将事件流按时间窗口划分为事件流片段,然后提取各时间窗口内的事件流片段特征,同时增强事件数量较多的时间窗口内特征。并基于奖励调节STDP规则帮助网络学习诊断性特征。采用增强算法的网络在MNIST-DVS数据集上的分类精度优于未采用增强算法的网络,并且对于较短的事件流输入也有很好的分类能力。该事件流特征增强算法能够提高基于Gabor滤波器的事件驱动卷积对事件相机对象的特征提取能力。Gabor filter-based event-driven convolution is a widely used feature extraction method in bio-inspired hierarchical spiking neural networks for event camera object recognition.To enhance the feature extraction performance of event camera objects,a Gabor filter-based event stream feature enhancement algorithm was proposed and applied to a spiking neural network event camera object recognition system that used reward-modulated STDP rule.The proposed algorithm first segmented event streams into fixed-length fragments,then extracted features from each fragment and enhanced features in fragments with high event density.The system used reward-modulated STDP rule to help the spiking neural network learn diagnostic features.The proposed algorithm outperformed the spiking neural network without the event stream feature enhancement algorithm on the MNIST-DVS dataset,and can classify short event stream input data effectively.The proposed event stream feature enhancement algorithm improves the feature extraction and classification performance of Gabor filter-based event-driven convolution for event camera object recognition.

关 键 词：事件相机 对象识别 特征增强 GABOR滤波器 奖励调节STDP 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]