lenet5 matlab

LeNet-5 is a convolutional neural network (CNN) architecture that was proposed by Yann LeCun et al. in 1998. It was designed to recognize handwritten digits in images and was used for the MNIST digit classification task. The LeNet-5 architecture consists of seven layers:

1. Input layer: receives the input image (28 x 28 pixels)
2. Convolutional layer: applies a set of learnable filters to the input image to extract features
3. Pooling layer: reduces the dimensionality of the feature maps by subsampling
4. Convolutional layer: applies a second set of filters to the feature maps
5. Pooling layer: further reduces the dimensionality of the feature maps
6. Fully connected layer: maps the high-level features to the output classes
7. Output layer: produces the final classification result

To implement LeNet-5 in MATLAB, one can use the Deep Learning Toolbox. The following code provides an example of how to define and train a LeNet-5 model for the MNIST dataset:

% Load the MNIST dataset
[XTrain, YTrain, XTest, YTest] = digitTrain4DArrayData;

% Define the LeNet-5 model architecture
layers = [
    imageInputLayer([28 28 1])
    convolution2dLayer(5, 6)
    tanhLayer
    averagePooling2dLayer(2, 'Stride', 2)
    convolution2dLayer(5, 16)
    tanhLayer
    averagePooling2dLayer(2, 'Stride', 2)
    fullyConnectedLayer(120)
    tanhLayer
    fullyConnectedLayer(84)
    tanhLayer
    fullyConnectedLayer(10)
    softmaxLayer
    classificationLayer];

% Specify the training options
options = trainingOptions('sgdm', ...
    'MaxEpochs', 20, ...
    'MiniBatchSize', 128, ...
    'InitialLearnRate', 0.01, ...
    'Shuffle', 'every-epoch', ...
    'Verbose', false, ...
    'Plots', 'training-progress');

% Train the model
net = trainNetwork(XTrain, YTrain, layers, options);

% Evaluate the model on the test set
YPred = classify(net, XTest);
accuracy = sum(YPred == YTest) / numel(YTest);
disp(['Test accuracy: ' num2str(accuracy)]);

This code defines a LeNet-5 model with the same architecture as the original paper, except that ReLU activation functions are replaced with hyperbolic tangent (tanh) functions. The model is trained using stochastic gradient descent with momentum (SGDM) and achieves a test accuracy of around 98%.