摘要
针对传统英语翻译机器人在手势识别中识别准确率低,导致人机交互效果不佳的问题,在传统卷积神经网络Lenet-5网络模型的基础上,对其卷积层、池化层连接层数和连接方式进行改进;然后构建基于改进的Lenet-5神经网络的手势识别模型,并通过损失函数和指数衰减、Adam算法和归一化算法分别对模型进行训练,以降低模型梯度,提升手势识别速度和精度。实验结果表明,提出算法的手势识别准确率高达99.6%,分别高出传统的Lenet-5网络模型和SVM分类7%和6.5%,且将本算法应用到英语翻译机器人的交互中发现,该机器人能够实现手势识别,智能交互能力显著提升。由此证明,提出的方法能够提升手势识别准确率和人机交互效果,可在智能英语翻译机器人中进行应用和推广。
For the problem of low recognition accuracy in gesture recognition,the convolutional layer and the traditional convolutional neural network Lenet-5 network model are improved by improving Lenet-5 neural network,and train the model through loss function and exponential attenuation,Adam algorithm and normalization algorithm to reduce the model gradient and improve the speed and accuracy of gesture recognition.The experimental results show that the gesture recognition accura-cy of the proposed algorithm is as high as 99.6%,7%and 6.5%higher than the traditional Lenet-5 network model and SVM classification,respectively.The application of this algorithm to the interaction of the English translation robot is found that the robot can realize gesture recognition,and the intelligent interaction ability is significantly improved.This result proves that the proposed method can improve the accuracy of gesture recognition and the human-computer interaction effect,and can be ap-plied and promoted in the intelligent English translation robot.
作者
张冠萍
ZHANG Guanping(Xi’an siyuan University,Xi’an 710038,China)
出处
《自动化与仪器仪表》
2022年第10期192-196,共5页
Automation & Instrumentation
基金
陕西省教育科学规划课题《幼儿园混龄教育的现状与效果研究》(SGH17H448)。
关键词
手势识别
翻译机器人
Lenet-5网络模型
人机交互
优化算法
gesture recognition
translation robot
Lenet-5 network model
human-computer interaction
optimization algorithm
作者简介
张冠萍(1982-),女,陕西西安人,本科,讲师,主要研究方向为英语教育技术方向。
