基于CNN-BiLSTM-SA网络的人类活动识别  

Human activity identification based on CNN-BiLSTM-SA network

作　　者：王赛[1] 张立新[1,2] 陈乃源 阚希 王军昂 吴凯枫 WANG Sai;ZHANG Lixin;CHEN Naiyuan;KAN Xi;WANG Junang;WU Kaifeng(School of Automation,Nanjing University of Information Science&Technology,Nanjing 210044,China;Wuxi University,Wuxi 214105,China)

机构地区：[1]南京信息工程大学自动化学院,江苏南京210044 [2]无锡学院,江苏无锡214105

出　　处：《西安理工大学学报》2024年第2期253-259,290,共8页Journal of Xi'an University of Technology

基　　金：国家自然科学基金资助项目(42105143);江苏省教育厅,江苏省高等学校基础科学(自然科学)研究资助项目(580221016)。

摘　　要：针对传统的神经网络对人类活动行为识别精度不高的问题,本文提出了一种基于双通道机制的卷积神经网络叠加双向长短期记忆网络和自注意力的混合网络模型(convolutional neural network-bi-directional long short-term memory-self-attention,CNN-BiLSTM-SA)。首先将数据集中的加速度和角速度数据作为网络的两个输入,然后使用卷积神经网络叠加双向长短期记忆网络的模式搭建系统,最后引入自注意力机制增强系统的分类能力。实验结果表明,在UCI-HAR数据集中,本网络的平均F 1分数达到98.6%,平均准确率达到98.4%,比卷积神经网络叠加长短期记忆神经网络模型(convolutional neural network-long short-term memory,CNN-LSTM)收敛速度更快并且准确率提高了4.2%,在人类活动识别方面,具有更广阔的应用前景。In view of the problem that the traditional neural network is not accurate in recognizing human activities,this paper proposes a hybrid network model based on the two-channel mechanism of convolutional neural network superimposed with bi-directional long short-term memory network and self-attention(CNN-BiLSTM-SA).First,the acceleration and angular velocity data in the data set are used as the two inputs of the network,and then the system is built by using the convolution neural network to overlay the bidirectional short-term and short-term memory network;finally,the self-attention mechanism is introduced to enhance the classification ability of the system.The experimental results show that in the UCI-HAR data set,the average F 1 score of this network is 98.6%,and that the average accuracy is 98.4%,which is faster than the convolutional neural network-long short-term memory(CNN-LSTM)convergence speed with the accuracy increased by 4.2%,and having a broader application prospect in human activity recognition.

关 键 词：人类活动识别 传感器 CNN-BiLSTM 自注意力机制 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置]