[handwritten numeral recognition] handwritten numeral recognition based on lenet network with matlab code

1 Content introduction

In today's society , Artificial intelligence has developed rapidly , And pattern recognition   As an important application field of artificial intelligence, it has also been flying   Rapid development , It uses the computer to calculate according to the   Features classify samples , Among them, optical character recognition technology is subject to   To the favor of the majority of research scholars . Handwritten numeral recognition is optical character   A branch of symbol recognition technology , It mainly studies how to use electronic meters   The computer automatically recognizes Arabic numerals written by people on paper . Present stage   The main research methods of segment handwritten numeral recognition are statistics 、 Clustering analysis   And neural networks .  Minimum distance classification algorithm is more traditional in pattern recognition 、 Simple   One of the ways , But the adaptability to handwritten Fonts is not high ［1］.K  most   Nearest neighbor algorithm （KNN） It is a classification algorithm based on statistics , most   Before 1968 Year by year Cover and Hart Put forward , As the simplest machine   Learning methods of learning , Theoretical research has been relatively mature ［2］. Support vector   machine （SVM） By Corinna Cortes and Vapnik be equal to 1995 year   First of all , The algorithm has good generalization ability and learning ability   force , The goal is to minimize structural risk , The obtained solution is global   Optimal solution , Overcome “ Dimension disaster ” problem , Greatly increase efficiency ［3］.BP（back propagation） Neural network is  1986  Year by year  Rumel⁃ hart  and  McCelland  Led by a team of scientists , As   A multilayer feedforward network , It is established on the basis of error inverse propagation   Of ［4］, It is often used in combination with optimization methods such as gradient descent method .BP  Neural network algorithm is a supervised training method , And activation letter   The number is nonlinear differentiable ［5］. These traditional recognition methods   The mathematical function representation ability of complex classification problems and network   The generalization ability is limited , It often fails to meet the requirements of high-precision recognition ［6］.  Convolutional neural networks （CNN） Originally by American scholars Cun Etc ,  It is a deep neural network with local connection between layers ［7］, Need to be   Through the forward and reverse transmission of information . because  CNN  Partial connection of   Pick up 、 Weight sharing and pooling operation ,CNN  It can effectively reduce   Reduce the complexity of the network and the number of training parameters ［8］.  In recent years , Because neural network has generalization ability 、 memory 、  Nonlinear and self-learning ability and high-speed computing ability ［9］, So roll   Product neural network is widely used in handwritten digit recognition .LeNet-5 model   As a typical convolutional neural network used to recognize numbers   Constantly optimized and improved . This topic is in  LeNet-5  Based on the model , utilize  MNIST Character library , By changing the sample training method , So as to obtain a more efficient and accurate handwritten digital information automatic   Recognition method .

2 Simulation code

<span style="color:#333333"><span style="background-color:rgba(0, 0, 0, 0.03)"><code>function layer1_mp = max_pool( layer1 )</code><code>​</code><code>[H,W,C] = size(layer1);</code><code>​</code><code>col_odd = 1:2:W;</code><code>col_even = 2:2:W;</code><code>row_odd = 1:2:H;</code><code>row_even = 2:2:H;</code><code>​</code><code>f1 = layer1(row_odd,col_odd,:);</code><code>f2 = layer1(row_odd,col_even,:);</code><code>f3 = layer1(row_even,col_odd,:);</code><code>f4 = layer1(row_even,col_even,:);</code><code>​</code><code>layer1_mp = zeros(size(f4));</code><code>​</code><code>for c = 1:C</code><code>    max_val1 = max(f1(:,:,c),f2(:,:,c));</code><code>    max_val2 = max(f3(:,:,c),f4(:,:,c));</code><code>    max_val = max(max_val1,max_val2);</code><code>    layer1_mp(:,:,c) = max_val;</code><code>end</code><code>end</code><code>​</code><code>​</code></span></span>

3 Running results

4 reference

[1] Mei Yanyu , Liao Qian , Shaowanling . be based on LeNet Research on handwritten digit recognition technology based on Neural Network [J]. New technology and process , 2020(6):3.

[2] Ding Jiao , Cong Shuang , Lin Wenrui , etc. . MATLAB In the environment LeNet Handwritten digit recognition on the Internet . 

About bloggers ： Good at intelligent optimization algorithms 、 Neural networks predict 、 signal processing 、 Cellular automata 、 The image processing 、 Path planning 、 UAV and other fields Matlab Simulation , relevant matlab Code problems can be exchanged by private letter .

Some theories cite network literature , If there is infringement, contact the blogger to delete .