我无法识别使用CNN的多字符输入图像，该图像使用自己的字符图像数据集进行了训练

Question

我使用自己的角色图像数据集训练了CNN，这是我自己创建的。然后我遵循https://in.mathworks.com/help/deeplearning/examples/create-simple-deep-learning-network-for-classification.html。到目前为止，我给出了90％以上的准确性。 在上面提到的过程中，它只是作为训练和测试从数据集中分为两部分。 但是，我想训练所有的数据集而不拆分，并且想识别包含数字单词的输入图像中的字符。 我试着做，但结果令人惊讶。 CNN培训代码：

close all;
clear all;
clc;
%%%%-------Load and Explore Image Data----%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
digitDatasetPath = fullfile('abcd');
imds = imageDatastore(digitDatasetPath, ...
'IncludeSubfolders',true,'LabelSource','foldernames');
labelCount = countEachLabel(imds);
numTrainFiles = 180;
imdsTrain = splitEachLabel(imds,numTrainFiles,'randomize'); %%%%%% I Used 
all images for training
layers = [
imageInputLayer([28 28 1])

convolution2dLayer(3,8,'Padding','same')
batchNormalizationLayer
reluLayer

maxPooling2dLayer(2,'Stride',2)

convolution2dLayer(3,16,'Padding','same')
batchNormalizationLayer
reluLayer

maxPooling2dLayer(2,'Stride',2)

convolution2dLayer(3,32,'Padding','same')
batchNormalizationLayer
reluLayer
fullyConnectedLayer(10)
softmaxLayer
classificationLayer];
options = trainingOptions('sgdm', ...
'InitialLearnRate',0.01, ...
'MaxEpochs',4, ...
'Shuffle','every-epoch', ...   
'ValidationFrequency',30, ...
'Verbose',false, ...
'Plots','training-progress');
mynet = trainNetwork(imdsTrain,layers,options);
save mynet;

预处理和识别程序：

close all;
clear all;
clc;
%outputfolder='L:\cnn';
load mynet
aa=imread('0.tif');
% figure;
% imshow(a);
% title('Input Image');
bb=rgb2gray(aa);
% figure;
% imshow(b);
% title('gray image');
cc=imbinarize(bb);
%figure;
%imshow(cc);
% title('binary Image');
dd=~cc;
% figure;
%imshow(dd);
[x,y]=size(cc);
ee=sum(cc,2);
mat2=y-ee;
mat3=mat2~=0;
mat4=diff(mat3);
index1=find(mat4);
[q,w]=size(index1);%size of index2 matrix is q*w
kap=1;
lam=1;
while kap<((q/2)+1)%number of loops=number of lines
k=1;
mat5=([]);
   for j=(index1(lam)+1):1:index1(lam+1)
     mat5(k,:)=cc(j,:); %store the line segmented matrix
       k=k+1;
   end
   lam=lam+2;   
   kap=kap+1;
   %figure, imshow(mat5);
   %imsave();


   %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
   %%%%%%%% WORD SEGMENT%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
   %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
   ccc=imbinarize(mat5);
   [xx,yy]=size(ccc);
   eee=sum(ccc,1);
   mat22=xx-eee;
   mat33=mat22~=0;
   mat44=diff(mat33);
   index11=find(mat44);
   [qq,ww]=size(index11);%size of index2 matrix is q*w
   kap1=1;
   lam1=1;
   while kap1<((ww/2)+1)%number of loops=number of lines
         kk=1;
         mat55=([]);
         for jj=(index11(lam1)+1):1:index11(lam1+1)
             mat55(:,kk)=ccc(:,jj); %store the line segmented matrix
             kk=kk+1;
         end
            lam1=lam1+2;   
            kap1=kap1+1;
            %figure, imshow(mat55);


            %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
            %%%%%%   REMOVE UNWANTED SPACE     %%%%%%%
            %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
             invmat55=~mat55;
             chasum=sum(invmat55,2);
             measurements = regionprops(chasum == 0, 'Area', 
              'PixelIdxList');
             fiveLongRegions = find([measurements.Area] >= 4);
             theIndexes = 
                vertcat(measurements(fiveLongRegions).PixelIdxList);

             cccc=mat55;
             cccc(theIndexes,:)=0;
             bbbb=cccc;
             %figure, imshow(bbbb);

             measurements = regionprops(bbbb, 'Area', 'BoundingBox');
             allAreas = [measurements.Area];
            % Crop out each word
            a=001;
            for blob = 1 : length(measurements)
                  % Get the bounding box.
                  thisBoundingBox = measurements(blob).BoundingBox;
                  % Crop it out of the original gray scale image.
                  thisWord = imcrop(mat55, thisBoundingBox);
                  RI2 = imresize(thisWord,[28 28]);                      
                  YPred = classify(mynet,RI2);
                  figure;
                  imshow(RI2);
                  label = YPred;
                  title(string(label));

            end


   end

end

我希望我的程序分为以下两部分，即培训和测试，如下面的波纹管模型所示：测试数据输入图像包含行数和单词数。