%%%??matlab實現LeNet-5
%%%??作者：xd.wp
%%%??時間：2016.10.14??20:29
%%?程序說明
%??????????1、池化（pooling）采用平均2*2
%??????????2、網絡結點數說明：
%???????????????????????????輸入層：28*28
%???????????????????????????第一層：24*24（卷積）*6
%???????????????????????????第二層：12*12（pooling）*6
%???????????????????????????第三層：8*8（卷積）*16
%???????????????????????????第四層：4*4（pooling）*16
%???????????????????????????第五層：全連接40
%???????????????????????????第六層：全連接10
%??????????3、網絡訓練部分采用800個樣本，檢驗部分采用100個樣本
clear?all;clc;
%%?網絡初始化
layer_c1_num=6;
layer_c2_num=16;
%權值調整步進
yita=0.05;
bias=1;
%卷積核初始化
[kernel_c1kernel_c2]=init_kernel（layer_c1_numlayer_c2_num）;
%pooling核初始化
pooling_a=ones（22）/4;
%全連接層的權值
weight_full_1=rand（1640）/sqrt（40）;
weight_full_2=rand（4010）/sqrt（10）;
weight_c2=rand（616）/10;
weight_arr2num=rand（44layer_c2_num）/sqrt（16）;
disp（‘網絡初始化完成......‘）;
%%?開始網絡訓練
disp（‘開始網絡訓練......‘）;
for?n=10:20
????for?m=0:9
????????%讀取樣本
????????train_data=imread（strcat（num2str（m）‘_‘num2str（n）‘.bmp‘））;
????????train_data=double（train_data）;
????????%?????????%?歸一化
????????%?????????train_data=train_data/sqrt（sum（sum（train_data.^2）））;
????????%標簽label設置
????????label_temp=-ones（110）;
????????label_temp（1m+1）=1;
????????label=label_temp;
????????for?iter=1:10
????????????%前向傳遞進入卷積層1
????????????for?k=1:layer_c1_num
????????????????state_c1（::k）=convolution（train_datakernel_c1（::k））;
????????????????%進入pooling1
????????????????state_s1（::k）=pooling（state_c1（::k）pooling_a）;
????????????end
????????????%進入卷積層2
????????????[state_c2state_c2_temp]=convolution_c2（state_s1kernel_c2weight_c2）;
????????????%進入pooling層2
????????????for?k=1:layer_c2_num
????????????????state_s2_temp1（::k）=pooling（state_c2（::k）pooling_a）;
????????????end
????????????%將矩陣變成數
????????????for?k=1:layer_c2_num
????????????????state_s2_temp2（1k）=sum（sum（state_s2_temp1（::k）.*weight_arr2num（::k）））+bias;
????????????????state_s2（1k）=1/（1+exp（-state_s2_temp2（1k）））;
%?????????????state_s2（1k）=sum（sum（state_s2_temp1（::k）.*weight_arr2num（::k）））;
????????????end
????????????%16個特征數，進入全連接層1
????????????state_f1=state_s2*weight_full_1;
????????????%進入全連接層2
????????????state_f2=state_f1*weight_full_2;
????????????%%?誤差計算部分
????????????Error=state_f2-label;
????????????Error_Cost=sum（Error.^2）;
????????????if（Error_Cost<1e-4）
????????????????break;
????????????end
????????????%%?參數調整部分
????????????[kernel_c1kernel_c2weight_c2weight_full_1weight_full_2weight_arr2num]=CNN_upweight1（Errortrain_data...
????????????????state_c1state_s1...
????????????????state_c2state_s2_temp1...
????????????????state_s2state_s2_temp2...
????????????????state_f1state_f2...
????????????????kernel_c1kernel_c2...
????????????????weight_c2weight_full_1...
????????????????weight_full_2weight_arr2numyitastate_c2_temp）;
????????????
????????end
????end
end
disp（‘網絡訓練完成，開始檢驗......‘）;
%%?檢驗部分
count_num=0;
for?n=10:15
????for?m=0

?屬性????????????大小?????日期????時間???名稱
-----------?---------??----------?-----??----
?????目錄???????????0??2018-11-30?16:47??CNN_LeNet_test\
?????文件????????4655??2016-10-21?21:25??CNN_LeNet_test\CNN_LeNet_main.m
?????文件????????5179??2016-10-17?16:20??CNN_LeNet_test\CNN_upweight.m
?????文件????????4729??2016-10-18?19:44??CNN_LeNet_test\CNN_upweight1.m
?????文件????????4478??2016-10-18?12:08??CNN_LeNet_test\CNN_upweight2.m
?????文件?????????298??2016-10-15?16:48??CNN_LeNet_test\convolution.m
?????文件?????????430??2016-10-15?19:49??CNN_LeNet_test\convolution_c2.m
?????文件?????????232??2016-10-21?21:23??CNN_LeNet_test\init_kernel.m
?????文件?????????292??2016-10-15?16:51??CNN_LeNet_test\pooling.m
?????目錄???????????0??2018-11-30?16:47??CNN_LeNet_test\test_image\
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_0.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_1.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_10.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_100.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_101.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_102.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_103.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_104.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_105.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_106.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_107.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_108.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_109.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_11.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_110.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_111.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_112.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_113.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_114.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_115.bmp
?????文件????????1862??2010-05-28?18:22??CNN_LeNet_test\test_image\0_116.bmp
............此處省略9900個文件信息