Miltary Target Classification in MSTAR Dataset using MATLAB

Image Processing

% Step 1: Load and Preprocess the Dataset

trainFolder = 'E:\Project\3 Vehicles\TRAIN DATA'; % Update with the path to the "train" folder

testFolder = 'E:\Project\3 Vehicles\TEST DATA'; % Update with the path to the "test" folder

% Load the training dataset and label based on folder names (train folder)

imdsTrainVal = imageDatastore(trainFolder, 'IncludeSubfolders', true, 'LabelSource', 'foldernames');

% Convert the images to grayscale (if not already grayscale) and resize to 128x128

imdsTrainVal.ReadFcn = @(x) imresize(imread(x), [128 128]);

% Split the dataset into 80% training and 20% validation

[imdsTrain, imdsValidation] = splitEachLabel(imdsTrainVal, 0.8, 'randomized');

% Step 2: Define the CNN Architecture

layers = [

imageInputLayer([128 128 1]) % Grayscale images of size 128x128 with 1 channel

% First convolutional layer and ReLU activation

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

batchNormalizationLayer

reluLayer

% First max pooling layer

maxPooling2dLayer(2, 'Stride', 2)

% Second convolutional layer and ReLU activation

convolution2dLayer(3, 32, 'Padding', 'same')

batchNormalizationLayer

reluLayer

% Second max pooling layer

maxPooling2dLayer(2, 'Stride', 2)

% Fully connected layer

fullyConnectedLayer(64)

reluLayer

% Output layer (3 classes: BMP2, BTR70, T72)

fullyConnectedLayer(3)

softmaxLayer

classificationLayer

];

% Step 3: Specify Training Options

options = trainingOptions('sgdm', ...

'MaxEpochs', 15, ... % Number of epochs

'MiniBatchSize', 32, ... % Mini-batch size

'InitialLearnRate', 0.001, ... % Learning rate

'ValidationData', imdsValidation, ...

'ValidationFrequency', 10, ... % Validate every 10 iterations

'Verbose', false, ...

'Plots', 'training-progress'); % Display training progress

% Step 4: Train the CNN

net = trainNetwork(imdsTrain, layers, options);

% Step 5: Evaluate the Model

YPred = classify(net, imdsValidation); % Predict the validation set

YValidation = imdsValidation.Labels; % True labels

% Calculate accuracy

accuracy = mean(YPred == YValidation);

disp(['Validation accuracy: ', num2str(accuracy * 100), '%']);

% Confusion matrix for performance evaluation

confusionchart(YValidation, YPred);

% Step 6: Test the Model (using the test folder)

% Load test images from the "test" folder

imdsTest = imageDatastore(testFolder, 'IncludeSubfolders', true, 'LabelSource', 'foldernames');

imdsTest.ReadFcn = @(x) imresize(imread(x), [128 128]);

% Classify the test set images

YPredTest = classify(net, imdsTest);

YTest = imdsTest.Labels; % True labels for test images

% Calculate test accuracy

testAccuracy = mean(YPredTest == YTest);

disp(['Test accuracy: ', num2str(testAccuracy * 100), '%']);

% Confusion matrix for test set performance evaluation

confusionchart(YTest, YPredTest);

% Step 7: Visualize Some Test Results

% Randomly select 9 images from the test set to visualize

numImagesToShow = 20; % Number of images to display

idx = randperm(length(imdsTest.Files), numImagesToShow); % Randomly select images

figure;

for i = 1:numImagesToShow

subplot(5, 4, i);

% Load and preprocess the image

img = readimage(imdsTest, idx(i));

imgResized = imresize(img, [128 128]); % Ensure resizing

% Classify the image

[label, score] = classify(net, imgResized);

% Display the image and the predicted label

imshow(img);

title(['Predicted: ', char(label), ' (', num2str(max(score) * 100, '%.2f'), '%)']);

end

% Step 8: Save the Trained Model

save('E:\Project\3 Vehicles\Custom CNN method\TrainedModel.mat', 'net'); % Save the trained model