• 1. A Hand-writing digit recognition Application base on Neural network and covolutional neural networkColleague of Computer Science Inner Mongolia University Hohhot , Inner Mongolia , P.R China Email: huangzhiqiang@aliyun.com
  • 2. NN digit recognition1. Background 2. Method neural network back-propagation 3. Experiment training and testing samples neural network structure results
  • 3. OCR (Optical Character Recognition) has become one of the important methods in gathering information and information transformation. Digit recognition has a promising feature in many fields in society, such as the car license plate recognition , postcode recognition, the statistics of report forms and financial report forms. So the research on the Digit recognition is important.background
  • 4. neural networkNeural network is a machine learning method. method
  • 5. Back PropagationBack propagation(BP) is a neural network learning algorithm. method
  • 6. EnvironmentHardware: CPU: Intel core i7 3630QM, 2.4GHZ Memory: 8G Byte Software: OS: windows 7 (64) IDE: vs2010 (C++) Experiment
  • 7. Experiment-Training and testing SamplesTraining and testing samples are from MNIST ,http://yann.lecun.com/exdb/mnist/ Every pic is a 28*28 dot matrix There are 10 pics in a sample, and there are 891 samples Experiment
  • 8. Neural network designOur neural network has 3 layers: LayerInput layerHidden layerOutput layerDimension14*14=19619610meaningAverage grey value of four neighbor dotsEmpirical value , about 1~1.5 times of the inputTarget numbers , I.e. 0,1,2,3….9Experiment
  • 9. neural network designBP neural network Experiment
  • 10. resultsTraining samples: Testing samples:Training itemsSample picsRight predictionRight ratioTraining timevalue7000688898.4%603 476msTesting itemstest picsRight predictionRight ratioTraining timevalue1900179194.2632%908ms(0.48ms/pic)Experiment
  • 11. 卷积神经网络基于人工神经网络 在人工神经网络前,用滤波器进行特征抽取 使用卷积核作为特征抽取器 自动训练特征抽取器(即卷积核,即阈值参数)
  • 12. 卷积卷积其实是一个图像处理核 卷积用于增强图像的某种特征
  • 13. 卷积的例子
  • 14. 子采样降低图像分辨率 减少训练维数 增强网络对大小变化的适用性
  • 15. 一般卷积神经网络的结构
  • 16. 我的卷积神经网络结构
  • 17. 实验效果
  • 18. 训练时间次数及准确率
  • 19. 问题1 准确率太低 2 准确率抖动厉害 3 单线程,训练速度太慢
  • 20. Thank you for listening!