Machine Learning for Additive Manufacturing

Code for image database collection

Figure 1: Flowchart of process carried out

Read video and initialise variables.

% Reading video
v = VideoReader('FullVideo.avi');
% Obtain number of frames in video
NumOfFrames = v.NumFrames;

% Set Up matrices
meanB = zeros(NumOfFrames, 1); % Matrix containing mean b* values
mean50B = zeros(50,1); % Matrix containing mean b* values of past 50 frames
minimum = zeros(NumOfFrames-50, 1); % Matrix containing minium values of past 50 frames
Identifying frames based on mean b* values.

% Initial for loop to obtain Mean b* of all frames
for frame = 1:NumOfFrames
  thisFrame = read(v,frame);
  cform = makecform('srgb2lab'); % Change frame from rgb to lab colour space
  LabFrame = applycform(im2double(thisFrame),cform);
  meanB(frame) = mean2(LabFrame(:,:,3)); % Obtain mean b* value
end

% For loop to obtain array of minimum values
for i = 1:(NumOfFrames - 50)
  mean50B = meanB(i:i+50);
  minimum(i) = min(mean50B); % Minimum value os past 50 frames mean b*
end

% For loop to write and compare mininum
for j = 7:(NumOfFrames-50)
    if minimum(j) < (minimum(j-6)/2) % Compare minimum values of current frame and 6th frame prior
      outputBaseFileName = sprintf('Frame %4.4d.png', (j+44)); % File name based on frame number
      outputFullFileName = fullfile('D:\', 'frames', outputBaseFileName); % Folder to save frames in
      Frame = read(v,j+44); % Read frame
      imwrite(Frame, outputFullFileName); % Save frame in folder indicated above
    end
end

Delete files from same cycle that are unnecessary.

d = dir('D:\frames\Not valuable frames'); % Access folder
filenames = {d.name};% Access all filenames
Pnum = 00; % Define variable for previous file number

for i = 1:numel(filenames)
  fn = filenames{i}; % Access file name of specific file
  % Find all files that contain space and find first 2 digits of frame
  [num, cnt] = sscanf(fn(find(fn == ' ', 1, 'last')+1:end-6), '%d');
  % If statement to delete files with the same first 2 digits
  if cnt == 1 && isequal(num,Pnum)
    delete(fullfile('D:\frames\Not valuable frames', fn));
  end
  Pnum = num;
end

Manually select coordinates for image cropping

% h = imshow('Frame 0431.png'); % Show first frame saved
hp = impixelinfo; % Show location of pointer in frame
set(hp,'Position',[5 1 300 20]); % Position of information of location
Save cropped images
d2 = dir('D:\frames') % Establis folder to analyse
filenames = {d2.name}; % Find all filenames in the folder

% For loop to save cropped images
for i = 3: numel(filenames) % For all files
fn = filenames{iB; % Get current file number
I = imread(fn); % Read frame of current file
P1 = imcrop(I,[435 342 760 60]); % Crop frame for Sample 1 coordinates
P2 = imcrop(I,[435 465 760 60]); % Crop frame for Sample 2 coordinates
P3 = imcrop(I,[411 663 760 60]); % Crop frame for Sample 3 coordinates

% Save cropped images to desired folder and with desired name, S1,S2,S3
outputBaseFileName = sprintf('P1 %12s.png', fn);
outputFullFileName = fullfile('D:\S1', outputBaseFileName);
imwrite(P1, outputFullFileName);
outputBaseFileName = sprintf('P2 %12s.png', fn);
outputFullFileName = fullfile('D:\S2',outputBaseFileName);
imwrite(P2, outputFullFileName);
outputBaseFileName = sprintf('P3 %12s.png', fn);
outputFullFileName = fullfile('D:\S3', outputBaseFileName);
imwrite(P3, outputFullFileName);
end

Published with MATLAB® R2020b

Code for Image Classification

Figure 2: Flowchart of process carried out

Establish datasets using ImageDataStore function.

File = fullfile('C:\Training and Test Set');
DB = imageDatastore(File,'IncludeSubfolders', true, 'LabelSource','foldernames');

Display class names and counts.

Table = countEachLabel(DB)
Categories = tbl.Label;

Partition 336 images for training and 228 for testing.

Ilocation = fileparts(DB.Files{1});
imgSet = imageSet(strcat(Ilocation,'\..'),'recursive'); % Recursively scan entire image set folder
[training_set,test_set] = imgSet.partition(113); % Create training and test set
test_set = test_set.partition(76)

Create visual vocabulary.

tic % Start timer
% Use bagOfFeatures function to create vocabulary
% Set vocabulary size, type of feature point selection
% and amount of features to select from (Final was 7, grid, 80%)
BoVW = bagOfFeatures(training_set, 'VocabularySize',7,'PointSelection','Grid','StrongestFeatures', 0.8);
% Create array of presence of visual words on each image
imgdata = double(encode(BoVW, tr_set));
toc % Stop timer
return; % Stop running code

Visualise/Plot histograms of feature vectors.

% Plot histogram of random S1 image
img = read(training_set(1), randi(training_set(1).Count));
featureVector = encode(BoVW, img); % Encode data from BoVW
subplot(1,3,1); % Establish 1x3 image
bS1 = bar(featureVector, 'FaceColor', [0 0.7 0]);
titS1 = title({'Visual Word Occurrences of','Image in S1 Training Set'});
xlabS1 = xlabel('Visual Word Index');
ylabS1 = ylabel('Frequency ()');

% Set font and size
set(gca, 'FontName', 'Times New Roman')
set([xlabS1,ylabS1], 'FontSize', 13)
set([titS1], 'FontSize', 15)

% Plot histogram of random S2 image
img = read(tr_set(2), randi(tr_set(2).Count));
featureVector = encode(BoVW, img);
subplot(1,3,2);
bS2 = bar(featureVector, 'FaceColor', [0 0.5 0]);
titS2 = title({'Visual Word Occurrences of','Image in S2 Training Set'});
xlabS2 = xlabel('Visual Word Index');
ylabS2 = ylabel('Frequency (%)');

% Set font and size
set(gca, 'FontName', 'Times New Roman')
set([xlabS2,ylabS2], 'FontSize', 13)
set([titS2], 'FontSize', 15)

% Plot histogram of random S3 image
img = read(tr_set(3), randi(tr_set(3).Count));
featureVector = encode(BoVW, img);
subplot(1,3,3);
bS3 = bar(featureVector, 'FaceColor', [0 0.3 0]);
titS3 = title({'Visual Word Occurrences of','Image in S3 Training Set'});
xlabS3 = xlabel('Visual Word Index');
ylabS3 = ylabel('Frequency (%)');

% Set font and size
set(gca, 'FontName', 'Times New Roman')
set([xlabS3,ylabS3], 'FontSize', 13)
set([titS3], 'FontSize', 15)

Create a 339x8 table using the encoded features

ImageData = array2table(imgdata); % Convert array created earlier to table
% Link each category with each feature vector
Type = categorical(repelem({training_set.Description}', [training_set.Count], 1));
ImageData.Type = Type;

Use the image data to train a model and assess its performance using Classification learner Open Classification Learner and import trained model

classificationLearner

Test out accuracy on testing set

tic % Start timer
testImData = double(encode(BoVW, test_set));
testImData = array2table(testImData,'VariableNames',M0780.RequiredVariables); % Implement imported model
actualImType = categorical(repelem({test_set.Description}', [test_set.Count], 1)); % Obtain actual category

predictedOutcome = M0780.predictFcn(testImData); % Implement imported model M0780 to test

% Count correct predictions by comparing obtained classification and actual category
correctPredictions = (predictedOutcome == actualImType);
validationAccuracy = sum(correctPredictions)/length(predictedOutcome) % Calculate validation accuracy
toc % Stop timer

Published with MATLAB® R2020b