基于自注意力卷积神经网络的吸烟行为识别方法  

Smoking Behavior Recognition Method Based on Self-Attention Convolutional Neural Network

作　　者：方武[1,2] 连圣阳 落莉莉 李慧姝 FANG Wu;LIAN Shengyang;LUO Lii;LI Huishu(Suzhou Institute of Trade and Commerce,Suzhou 215009,China;Suzhou L.J.T Intelligent Technology Co.,Ltd.,Suzhou 215004,China)

机构地区：[1]苏州经贸职业技术学院,江苏苏州215009 [2]苏州朗捷通智能科技有限公司,江苏苏州215004

出　　处：《现代信息科技》2024年第22期20-24,共5页Modern Information Technology

基　　金：江苏省高等学校基础科学(自然科学)研究面上项目(22KJB140014)。

摘　　要：卷积神经网络因其强大的特征提取能力,已成为当前视觉行为识别任务中的主流方法。为应对公共场所吸烟行为有效监测预警,文章提出了一种基于自注意力卷积神经网络的吸烟行为识别方法。通过分析吸烟行为在图像和视频中的关键特征,设计了一种高效的卷积神经网络模型。该模型通过引入自注意力机制,能准确高效提取图像的关键特征,以实现对吸烟行为的准确识别。实验结果表明,所提出的方法在不同场景下均表现出良好的识别效果和鲁棒性,具有较高的实用价值。Convolutional Neural Network has become the mainstream method in the current visual behavior recognition task because of its powerful feature extraction ability.In order to effectively monitor and warn smoking behavior in public places,this paper proposes a smoking behavior recognition method based on Self-Attention Convolutional Neural Network.By analyzing the key features of smoking behavior in images and videos,an efficient Convolutional Neural Network model is designed.This model can accurately and efficiently extract the key features of images by introducing Self-Attention mechanism to achieve accurate recognition of smoking behavior.The experimental results show that the proposed method has good recognition effect and robustness in different scenarios,and has high practical value.

关 键 词：卷积神经网络 吸烟行为识别 自注意力 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]