摘要
手语是正常人与听障人士的沟通桥梁,而动态手语是其最为重要的分支,在现实生活中有着重要的研究意义。针对动态手语识别中识别难度大、识别率低等问题,对手语数据集进行提取关键帧等预处理操作,使用深度学习的方法,以降维后的3DResNet卷积神经网络(Convolutional Neural Network,CNN)为基础,构建网络基础模型。实验采用迁移学习方法预训练模型,加快了网络的优化过程,并对其引入注意力机制以增强对手部信息的关注,提升了识别精度,其在CSL100手语数据集上的验证识别准确率达到95.11%,与其他方式对比,该方法具有较高的识别率,验证了此改进算法的有效性与可行性。
Sign language serves as a bridge between normal people and hearing-imparred person with communication difficulties,and dynamic sign language is one of the most important branches which have important research significance in real life.To address the issues of high difficulty and low recognition rate in dynamic sign language recognition,this paper first performs preprocessing operations such as extracting key frames on the sign language dataset.Subsequently,using deep learning methods,a network base model is constructed based on the dimensionality-reduced 3D ResNet Convolutional Neural Network(CNN).Furthermore,the experiment adopts transfer learning to pre-train the model,accelerating the network optimization process.Attention mechanisms are also introduced to enhance the focus on hand information,thereby improving recognition accuracy.Finally,the recognition accuracy of 95.11%is verified on the CSL100 sign language dataset.By comparing with other methods,this experimental approach demonstrates a higher recognition rate,thus validating the effectiveness and feasibility of the proposed improved algorithm.
作者
王特
郭莹
吴春迪
WANG Te;GUO Ying;WU Chundi(School of Information Science and Engineering,Shenyang University of Technology,Shenyang 110870,China)
出处
《计算机与网络》
2025年第6期609-615,共7页
Computer & Network
