resnet18 adapt to 1-d input in matlab

main.m
inputSize = 512; % size of the 1-D input
numClasses = 10; % number of output classes

% Load pre-trained ResNet-18
net = resnet18();

% Remove the last layer
lgraph = layerGraph(net);
lgraph = removeLayers(lgraph, {'fc1000', 'prob', 'ClassificationLayer_predictions'});

% Add a 1-D convolutional layer as the new input layer
newLayers = [
    imageInputLayer([1 inputSize 1], 'Name', 'input')
    convolution2dLayer([1 7], 64, 'Stride', [1 2], 'Padding', [0 3], 'Name', 'conv1')
    batchNormalizationLayer('Name', 'bn_conv1')
    reluLayer('Name', 'relu_conv1')
    averagePooling2dLayer([1 3], 'Stride', [1 2], 'Padding', [0 1], 'Name', 'pool1')
];
lgraph = addLayers(lgraph, newLayers);

% Connect the new layers to the existing layers
lgraph = connectLayers(lgraph, 'input', 'conv1');

% Add the new classification layers
newLayers = [
    fullyConnectedLayer(numClasses, 'Name', 'fc10')
    softmaxLayer('Name', 'softmax')
    classificationLayer('Name', 'classoutput')
];
lgraph = addLayers(lgraph, newLayers);
lgraph = connectLayers(lgraph, 'pool5', 'fc10');

% Plot the modified ResNet-18 architecture
plot(lgraph);

% Freeze the weights of the earlier layers and train only the new layers
net = trainNetwork(XTrain, YTrain, lgraph, options);
1252 chars
38 lines