首页|基于自适应注意力机制的深度学习人脸识别系统研究

基于自适应注意力机制的深度学习人脸识别系统研究

扫码查看
人脸识别已经成为社会生活、商业和公共服务中不可或缺的一部分,由于人脸图像具有复杂性和多样性的特点,其准确性和识别速率需要深入研究,本文旨在优化YOLOv5模型,构建一个快速、自动化且高效的人脸识别系统.首先在模型输入端增设特征库规范化功能,实现统一化以及最大化输入图片目标特征,接着设计自适应注意力机制以更好地关注不同尺寸的目标特征,优化损失函数提升整体性能,设计添加特征比对模块实现人脸识别,最后设计可视化界面增加系统的可操作性和客观性.模型训练使用Wider Face数据集,优化后模型召回率提升2.2%,mAP提升2.4%,高于原YOLOv5s模型.改进后的人脸识别系统,整体识别效果达到87.65%,为自动化会议考勤系统提供了有效方法.
Research on Deep Learning Face Recognition System Based on Adaptive Attention Mechanism
Facial recognition has become an indispensable part of social life,commerce and public services.Due to the com-plexity and diversity of facial images,their accuracy and recognition speed require in-depth research.This article aims to optimize the YOLOv5 model and build a fast,automated and efficient facial recognition system.Firstly,a feature library normalization func-tion is added to the input end of the model to achieve uniformity and maximization of input image target features.Then,an adaptive attention mechanism is designed to better focus on target features of different sizes,optimize the loss function to improve overall per-formance.A feature comparison module is designed to achieve facial recognition.Finally,a visual interface is designed to increase the maneuverability and objectivity of the system.The model is trained to use the Wider Face dataset,and after optimization,the recall rate of the model increased by 2.2%and mAP increased by 2.4%,which is higher than the original YOLOv5s model.The improved facial recognition system achieves an overall recognition effect of 87.65%,providing an effective method for automated conference attendance systems.

YOLOv5face recognitionattention mechanism

于晓、杨梦瑶

展开 >

天津理工大学电气工程与自动化学院,天津 300384

YOLOv5 人脸识别 注意力机制

2024

黑龙江工业学院学报(综合版)
鸡西大学

黑龙江工业学院学报(综合版)

影响因子:0.211
ISSN:1672-6758
年,卷(期):2024.24(9)