摘要
传统深度神经网络受到图形处理限制,运行速度较慢,难以对手势动作快速检测识别。为了解决这一问题,创建更加良好的人机交互智能应用,提出了基于单词多盒检测器(Single Shot MultiBox Detector,SSD)卷积神经网络的嵌入式手势检测算法。通过预处理方式,扩展原有数据库,并进一步分解神经网络卷积因子。在MobileNet神经网络的作用下提升运算速度,并在图形规格参数的变化下对网络结构加以优化,从而简化算法。经过测试证明该算法检测速度有所提升,尽管检测精度有所降低,但仍保持高精度标准。
Traditional deep neural networks are limited by graphics processing and run slowly,which makes it difficult to quickly detect and recognize gestures.In order to solve this problem and create a better human-computer interaction intelligent application,an embedded gesture detection algorithm based on SSD convolutional neural networks is proposed.By preprocessing,the original database is extended,and the convolutional factor of neural network is further decomposed.The operation speed is improved under the action of MobileNet neural network,and the network structure is optimized under the change of graph specification parameters,so as to simplify the algorithm.The test and verification show that the detection speed of the algorithm is improved,although the detection accuracy is reduced,it still maintains the high precision standard.
作者
杨宇超
YANG Yuchao(Tianjin Polytechnic College,Tianjin 300000,China)
出处
《信息与电脑》
2024年第11期30-33,共4页
Information & Computer
关键词
卷积神经网络
嵌入式
SSD
convolutional neural network
embedded
SSD
