How can I add additional features to a pretrained AlexNet?

I am classifying images using transfer learning and the pretrained AlexNet convolutional neural network.

I would like to add additional features in the fully connected layers at the end of the network.

Is it possible to add features that bypass the convolutional layers and are incorporated only at the fully connected layers?

Answer 1

[MathWorks Support Team]

You can do this by following the workflow:

1. Extract the features that is outputted by the pretrained  (& transfer learning) AlexNet network right before the fully connected layer:

https://www.mathworks.com/help/deeplearning/ug/extract-image-features-using-pretrained-network.html

2. Concatenate the features from step 1 with the additional features from your image metadata.

3. Create and train a fully-connected neural network that will take in the concatenated features from step 2.