摘要
支持向量机的手写体数字识别中,采用美国邮政服务数据库。并取多个2层神经网络中的最好者得出2层神经网络结果,专门设计5层卷积神经网络Lenetl。所有的结果均直接采用点阵输入,将像素值归正到相应区域间,且不施加任何预处理。该方法与人工分类、神经网络、决策树等方法比较,其测试误差低,测试速度高。
Handwritten number recognition based on support vector machine adopts the US post service database. Moreover, it calculate the two-layer nerve network results based on several best two-layer nerve networks and design five-layer convolution nerve networks Lenetl. All results are inputted by using lattice input. The image point value is sent to the corresponding area without pretreatment. Compared with artificial classification, nerve network, and decision tree, its test error is low and the speed is high.
出处
《兵工自动化》
2007年第3期39-41,共3页
Ordnance Industry Automation
关键词
支持向量机
手写体数字识别
卷积神经网络
SVM （Support Vector Machine）
Handwritten number recognition
Convolution nerve network
