摘要
近年,深度学习的发展使得手势识别卷积神经网络取得了突破性进展,但现有基于卷积神经网络的手势识别方法仍存在收敛速度慢、识别率低等问题,因此手势识别很难取得较好成果。通过对CNN训练过程中误差产生的原因及其反馈模型的分析,提出了一种自适应增强卷积神经网络(Adaptively Enhanced Convolution Neural Network,AE-CNN)的识别算法。结果表明,文中算法不仅实现了分类特征的自适应增强,同时也提高了收敛速度和识别率。
With the rapid development of deep learning,the convolution neural network( CNN) for gesture recognition has made a breakthrough in recent years. The existing methods based on CNN still have some problems such as slow convergence speed,low recognition rate,so it is difficult to achieve good results in gesture recognition. Against the slow convergence speed and low recognition rate of the CNN problems,this paper proposes an adaptively enhanced convolution neural network( AE-CNN)recognition algorithm. The results show that the proposed algorithm not only realizes the adaptive enhancement of classification features,but also improves the convergence speed and recognition rate.
作者
曹军梅
秦婧文
CAO Jun-mei;QIN Jing-wen(College of Computer Science and Technology, Yan’an University, Yan’an 716000,Shaanxi Province,China)
出处
《信息技术》
2019年第6期18-21,共4页
Information Technology
基金
国家自然科学基金项目(61761042)
延安大学科研引导项目(YDY2018-11)
国家级大学生创新训练计划项目(201710719017)
陕西省大学生创新训练计划项目(1531)