摘要
近年来,基于传感器手语识别研究受到广泛关注,而传统的手语分析方法很大程度上依赖于专家经验和信号处理技术的先验知识。为此,本文提出一种CGR_ATT模型自动提取手语特征。首先,构建卷积神经网络(CNN)作为特征提取器通过多层卷积堆叠和池化操作,从传感器数据中提取空间特征,无需先验知识即可从数据中识别并提取关键的空间信息;其次,通过门控循环单元(GRU)构建时序建模器,捕捉和处理空间信息在时间维度上的变化,利用门控机制处理传感器数据中的时序关系;最后,通过ATT模块给序列中的每个元素分配不同权重,进一步处理GRU输出的时序特征。所提模型较CLA_net、CLT_net、长短期记忆(LSTM)、GRU模型具有较高的预测精度,预测精度达到96.98%。
In recent years,research on sign language recognition using sensors has received widespread attention.However,traditional methods for sign language analyzing rely on expert experience and prior knowledge of signal processing techniques to a large extent.Therefore,a CGR_ATT model the automatic extraction of sign language features is proposed.Firstly,the convolutional neural network(CNN)is constructed as feature extractor.Through multiple layers of convolution and pooling operations,spatial features are extracted from sensor data without requiring prior knowledge,thereby identifying and extracting key spatial information directly from the data.Secondly,a time series modeler is constructed through GRU to capture and process the changes of spatial information in time dimension,and the gating mechanism is used to deal with the time series relationship in sensor data.Finally,the ATT module assigns different weights to each element in the sequence,and further processes the timing features of the GRU output.The proposed model has higher prediction precision compared to CLA_net,CLT_net,long short-term memory(LSTM),and GRU models,achieving a prediction precision of 96.98%.
作者
武东辉
王金凤
仇森
刘国志
WU Donghui;WANG Jinfeng;QIU Sen;LIU Guozhi(College of Building Environment Engineering,Zhengzhou University of Light Industry,Zhengzhou 450000,China;School of Control Science and Engineering,Dalian University of Technology,Dalian 116081,China)
出处
《传感器与微系统》
北大核心
2025年第2期21-25,30,共6页
Transducer and Microsystem Technologies
基金
国家自然科学基金资助项目(62272081)
河南省科技公关项目(222102210086,242102220099,232102320046,232102321021)
河南省高等学校重点科研项目(25B413005)。
作者简介
通讯作者:武东辉(1984-),男,博士,硕士研究生导师,主要研究领域为机器学习、模式识别;王金凤(1999-),女,硕士研究生,研究方向为智慧感知与行为分析、深度学习。
