计算机与数字工程2025，Vol.53Issue(4)：1127-1131,1163,6.DOI:10.3969/j.issn.1672-9722.2025.04.036
基于注意力机制的P3D卷积神经网络人体行为识别
Attention-based P3D Convolutional Neural Network Human Action Recognition
摘要
Abstract
Since the attention mechanism can effectively extract the key information in the video frame,the introduction of the attention mechanism on the P3D model helps to enhance the key features in human action recognition.This paper proposes a P3D ac-tion recognition method based on attention mechanism(CBAM-P3D).The CBAM attention mechanism introduced by this method in the P3D model includes two modules,which are spatial attention(SAM)and channel attention(CAM).CAM mainly obtains the key frames of the video by focusing on the channel of each frame of the video and between two frames.SAM mainly focuses on the key spatial features on each frame.The model in this paper is tested on the UCF101 and HMDB51 datasets respectively.Experi-ments show that the recognition accuracy of the model has been further improved.关键词
注意力机制/P3D模型/人体行为识别/数据集Key words
attention mechanism/P3D model/human action recognition/datasets分类
信息技术与安全科学引用本文复制引用
唐璇,王艳霞..基于注意力机制的P3D卷积神经网络人体行为识别[J].计算机与数字工程,2025,53(4):1127-1131,1163,6.基金项目
重庆市科委科学研究项目(编号:cstc2021jcyj-msxm2791) （编号:cstc2021jcyj-msxm2791）
重庆市教委科技项目(编号:KJZD-K202200513)资助. （编号:KJZD-K202200513）
