Classification of shapes by CNN

This is the last part. Previous posts were about transfer learning, and about creating set of images.

In the previous post we trained our CNN with high accuracy. This sample tests the network on 16 images that are placed in one .png file.

A = imread('shapes16.png');%read image with 16 shapes

%divide image into cells
A_in_cells  = mat2cell(A,[227 227 227 227],[227 227 227 227],[1 1 1]);

ind = 1;
for ii = 1:4 
    for yy = 1:4
        %compose cells to image
        one_shape = cat(3,A_in_cells{ii,yy,1},A_in_cells{ii,yy,2},A_in_cells{ii,yy,3});
        [label,score] = classify(netTransfer,one_shape);%classification
        Classified_images{ind} = insertText(one_shape,[1 1],... %add label
            cellstr(string(label) + " "+ num2str(max(score)*100)+" %"),'FontSize',26);
        ind = ind+1;
    end
end

montage(Classified_images)