摘要
残差网络作为卷积神经网络中的经典模型,受到了研究者的广泛关注,因此产生了多种衍生模型。同时,手势识别也是当前的热点研究领域,在利用残差网络实现手势识别方面已有大量研究成果。本文利用了多种残差网络模型的衍生模型,对ASL手势数据集进行训练,得到了不同模型下的实验结果。其中,训练结果最好的模型是Res Net18v1,它的识别正确率最高可达到93.3%。研究结果表明:在残差网络的衍生模型中,所堆叠的卷积层数越多,对准确率的提升效果不一定越强,需要根据任务要求,灵活选择模型并应用。
Residual network,as a classic model in convolutional neural networks,has received extensive attention from researchers,and a variety of derivative models have been produced.At the same time,gesture recognition is also a current hot research field,and there are a lot of research results in the use of residual network to realize gesture recognition.In this paper,a variety of derivative models of residual network models are used to train the ASL gesture data set,and experimental results under different models are obtained.The model with the best training result is Res Net18v1,and its recognition accuracy can reach 93.3%.The research results showthat in the derivative model of the residual network,more convolutional layers stacked do not equal better accuracy.Therefore,the model needs to be flexibly selected and applied according to the task requirements.
作者
郝禹哲
袁天夫
田海越
HAO Yuzhe;YUAN Tianfu;TIAN Haiyue(School of Electronic and Electrical Engineering,Shanghai University of Engineering Science,Shanghai 201620,China;School of Mechanical and Automotive Engineering,Shanghai University of Engineering Science,Shanghai 201620,China)
出处
《智能计算机与应用》
2020年第7期64-66,共3页
Intelligent Computer and Applications
基金
国家大学生创新项目(201910856009)
关键词
手势识别
卷积神经网络
残差网络
Gesture recognition
Convolutional neural network
Residual Network
