文献类型：期刊文献

中文题名：基于轻量级DenseNet和ZigBee的指纹识别方法

英文题名：Fingerprint Recognition Method Based on Lightweight DenseNet and ZigBee

作者：叶得学[1];韩如冰[1];颜鲁合[2]

第一作者：叶得学

机构：[1]兰州工商学院信息工程学院,甘肃兰州730101;[2]甘肃中医药大学经贸与管理学院,甘肃兰州730000

第一机构：兰州工商学院信息工程学院,甘肃兰州730101

年份：2023

卷号：38

期号：4

起止页码：78

中文期刊名：湖南科技大学学报（自然科学版）

外文期刊名：Journal of Hunan University of Science And Technology：Natural Science Edition

收录：CSTPCD;;北大核心:【北大核心2020】；

基金：甘肃省科技计划项目资助(20CX9ZA021,22JR5RA809,20CX9ZA068);兰州市科技计划项目资助(2019-ZD-167,2020-ZD-139)。

语种：中文

中文关键词：指纹识别;轻量级DenseNet;密集连接卷积神经网络;ZigBee;深度学习;

外文关键词：fingerprint recognition;lightweight DenseNet;densely connected convolutional neural network;ZigBee;deep learning

摘要：针对传统身份识别方法识别准确率低、模型复杂且运算速度慢等问题,提出一种新的基于轻量级密集连接卷积神经网络(DenseNet)和紫蜂协议(ZigBee)的指纹识别方法.首先,构建指纹识别系统的整体模型,并对该模型进行适当的裁剪以缩减模型复杂度.其次,通过筛选指纹图像、增强有效指纹以及扩充增强后的数据等操作,对采集到的ZigBee协议指纹信息进行预处理.然后,以传统深度残差网络的基本思想为依据,通过前馈的方式改变层间的连接关系并构建轻量级DenseNet.最后,以相同的样本数据为基础分别对轻量级DenseNet模型、普通DenseNet模型和3种传统基于机器学习的身份识别算法模型进行训练.试验结果表明:所提出的基于轻量级DenseNet模型的识别准确度最高,为98.24%,且该模型的运行速度最快,与普通DenseNet模型相比,其模型复杂度降低了94%以上.
With the gradual popularity of smart devices in daily life,traditional identification methods have low recognition accuracy,complex models,and slow calculation speeds.In response to the above problems,this paper proposes a fingerprint recognition method based on a lightweight densely connected convolutional neural network(DenseNet)and the ZigBee protocol(ZigBee).First of all,this paper theoretically constructs the overall model of the fingerprint recognition system,and appropriately tailors the model to reduce the complexity of the model.Secondly,through operations such as screening fingerprint images,enhancing effective fingerprints,and expanding the enhanced data,the collected fingerprint information of the ZigBee protocol is preprocessed.Then,based on the basic idea of the traditional deep residual network,the connection relationship between the layers is changed in a feedforward manner,and a lightweight DenseNet is constructed.Finally,based on the same sample data,the lightweight DenseNet model,the ordinary DenseNet model,and the other three traditional machine learning-based identity recognition algorithm models are trained respectively.The experimental results show that the average recognition accuracy of the proposed lightweight DenseNet model is 98.24%,and the model runs the fastest.Compared with the ordinary DenseNet,its model complexity is reduced by more than 94%.