how can i add feature map to the CNN before fullyConnectedLayer ?

layers = [

imageInputLayer([width,height,channels]); %'DataAugmentation', 'none'); %'Normalization', 'none');

convolution2dLayer(2,16,'Padding',1)

batchNormalizationLayer

reluLayer

here add feature map

fullyConnectedLayer(12)

softmaxLayer

classificationLayer];

Answer 1

[Shantanu Dixit]

Hi Amir,

To add a feature map in a neural network you can typically add more convolutional layers, which are responsible for detecting features in an image. In the context of your MATLAB code, you can add additional "convolution2dLayer" followed by "batchNormalizationLayer" and "reluLayer" to expand the feature map. Here's how you can modify your code:

% example height width and channels
width = 28; 
height = 28; 
channels = 1; 

%% To calculate the output size of each layer, you can use the formula:
%% output feature size = floor( (Input size + 2*(padding) - filter size) / stride) + 1
%% layer = convolution2dLayer(filterSize,numFilters,Name=Value)

layers = [
    imageInputLayer([width, height, channels])
    convolution2dLayer(2, 16, 'Padding', 1)
    batchNormalizationLayer
    reluLayer
    convolution2dLayer(3, 32, 'Padding', 1)
    batchNormalizationLayer
    reluLayer
    convolution2dLayer(3, 64, 'Padding', 1)
    batchNormalizationLayer
    reluLayer
    fullyConnectedLayer(12)
    softmaxLayer 
];

Additionally you can refer to the following MathWorks documentation for more information:

• 'convolution2dLayer': https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layer.convolution2dlayer.html
• 'batchNormalizationLayer': https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layer.batchnormalizationlayer.html

Hope this helps!